JP7264534B2 - Determination of base modifications of nucleic acids - Google Patents

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is filed May 4, 2020, U.S. Provisional Patent Application No. 63/051,210, entitled "Determination of Base Modifications of Nucleic Acids," filed July 13, 2020. U.S. Provisional Patent Application No. 63/019,790 entitled "Determination of Base Modifications of Nucleic Acids," U.S. Provisional Patent Application No. 62/019,790, entitled "Determination of Base Modifications of Nucleic Acids," filed March 19, 2020. 991,891, U.S. Provisional Patent Application No. 62/970,586, entitled "Determination of Base Modifications of Nucleic Acids," filed February 5, 2020; No. 62/887,987 entitled "Determination of Base Modifications of Nucleic Acids". The contents of all of these are hereby incorporated by reference for all purposes.

The presence of base modifications in nucleic acids varies in different organisms, including viruses, bacteria, plants, fungi, nematodes, insects, vertebrates (eg, humans), and the like. The most common base modification is the addition of methyl groups to different DNA bases at different positions, so-called methylation. Methylation is 5mC (5-methylcytosine), 4mC (N4-methylcytosine), 5hmC (5-hydroxymethylcytosine), 5fC (5-formylcytosine), 5caC (5-carboxylcytosine), 1mA (N1-methylcytosine) adenine), 3 mA (N3-methyladenine), 7 mA (N7-methyladenine), 3mC (N3-methylcytosine), 2mG (N2-methylguanine), 6mG (O6-methylguanine), 7mG (N7-methylguanine) , 3mT (N3-methylthymine), and 4mT (O4-methylthymine), such as cytosines, adenines, thymines, guanines. In vertebrate genomes, 5mC is the most common type of base methylation, followed by guanine methylation (ie, in the context of CpGs).

DNA methylation is essential for mammalian development, gene expression and silencing, embryonic development, transcription, chromatin structure, X-chromosome inactivation, protection against repetitive element activity, maintenance of genome stability during mitosis , as well as play a notable role in regulating genomic imprinting of parental origin.

DNA methylation plays many important roles in promoter and enhancer silencing in a coordinated manner (Robertson, 2005; Smith and Meissner, 2013). Many human diseases have been found to be associated with abnormal DNA methylation, including but not limited to carcinogenic processes, imprinting disorders (e.g. Beckwith-Wiedemann syndrome and Prader-Willi syndrome), recurrent Instability disorders (eg, fragile X syndrome), autoimmune disorders (eg, systemic lupus erythematosus), metabolic disorders (eg, type I and type II diabetes), neuropathies, aging, and the like.

Accurate measurement of methylomic modifications of DNA molecules has many clinical implications. One widely used method to measure DNA methylation is to use bisulfite sequencing (BS-seq) (Lister et al., 2009; Frommer et al., 1992). In this approach, a DNA sample is first treated with bisulfite to convert unmethylated cytosines (ie, C) to uracil. In contrast, methylated cytosines remain unchanged. The bisulfite-modified DNA is then analyzed by DNA sequencing. In another approach, following bisulfite conversion, the modified DNA is then subjected to polymerase chain reaction (PCR) amplification using primers that can distinguish bisulfite-converted DNA of different methylation profiles (Herman et al., 1996). This latter approach is called methylation-specific PCR.

One drawback of such bisulfite-based approaches is that the bisulfite conversion step has been reported to significantly degrade most of the treated DNA (Grunau, 2001). Another drawback is that the bisulfite conversion step produces a strong CG bias (Olova et al., 2018), typically resulting in a reduced signal-to-noise ratio for DNA mixtures with heterogeneous methylation states. That is. Furthermore, bisulfite sequencing cannot sequence long DNA molecules because the DNA is degraded during the bisulfite treatment. Therefore, there is a need to determine base modifications of nucleic acids without prior chemical treatment (eg, bisulfite conversion) and nucleic acid amplification (eg, using PCR).

We have found that in one embodiment, it enables the determination of base modifications such as 5mC in nucleic acids without enzymatic and/or chemical transformations or pretreatment of the template DNA such as protein and/or antibody binding. developed a new method to Such pretreatment of template DNA is not required for determination of base modifications, but in the examples shown, certain pretreatments (e.g., digestion with restriction enzymes) serve to enhance aspects of the invention. (eg allowing enrichment of CpG sites for analysis). Embodiments present in the present disclosure include, for example, but not limited to, 4mC, 5hmC, 5fC, and 5caC, 1 mA, 3 mA, 7 mA, 3 mC, 2 mG, 6 mG, 7 mG, 3 mT and 4 mT, and the like. can be used to detect Such embodiments take advantage of sequencing-derived features such as kinetic features that are affected by various base modifications, as well as the identity of nucleotides in the window around the target position in which methylation status is determined. can be done.

Embodiments of the present invention can be used for, but not limited to, single molecule sequencing. One type of single-molecule sequencing is single-molecule real-time sequencing, which monitors the sequencing progress of a single DNA molecule in real-time. One type of single-molecule real-time sequencing has been commercialized by Pacific Biosciences using single-molecule real-time (SMRT) systems. The method can use the pulse width of the signal from the sequencing bases, the interpulse duration (IPD) of the bases, and the identity of the bases to detect modifications of the bases or nearby bases. . Another single-molecule system is a system based on nanopore sequencing. One example of a nanopore sequencing system is that commercialized by Oxford Nanopore Technologies.

The method developed by the inventors detects base modifications in biological samples and serves as a tool to assess the methylation profile of samples for a variety of purposes, including but not limited to research and diagnostic purposes. The detected methylation profile can be used for different analyses. Methylation profiles can be used to detect the origin of DNA (eg, DNA obtained from maternal or fetal, tissue, bacteria, or tumor cells enriched from the blood of cancer patients). Detection of aberrant methylation profiles in tissues helps identify developmental disorders in individuals, identify and predict tumors or malignancies.

Embodiments of the invention may include analyzing relative methylation levels of haplotypes in an organism. The imbalance in methylation levels between the two haplotypes can be used to determine disorder classification. A larger imbalance may indicate the presence of a disability, or a more severe disability. Disorders can include cancer.

Chimeric and hybrid DNA can be identified by single-molecule methylation patterns. Chimeric and hybrid molecules can include sequences from two different genes, chromosomes, organelles (eg, mitochondria, nuclei, chloroplasts), organisms (mammals, bacteria, viruses, etc.), and/or species. Detecting junctions in chimeric or hybrid DNA molecules may allow detection of gene fusions in various disorders or diseases, including cancer, prenatal or congenital disorders.

A better understanding of the nature and advantages of embodiments of the present invention may be obtained with reference to the following detailed description and accompanying drawings.

Figure 3 shows SMRT sequencing of molecules with base modifications according to embodiments of the invention. Figure 3 shows SMRT sequencing of molecules with methylated and unmethylated CpG sites according to embodiments of the invention. 4 shows pulse intervals and pulse widths according to embodiments of the present invention; FIG. 10 shows an example of a Watson strand measurement window of DNA for detecting base modifications according to embodiments of the present invention. FIG. FIG. 10 shows an example of a DNA click strand measurement window for detecting base modifications according to embodiments of the present invention. FIG. FIG. 10 shows an example of a measurement window by combining data from the Watson strand of DNA and its complementary Crick strand for detecting arbitrary base modifications according to embodiments of the present invention. FIG. FIG. 11 shows an example of a measurement window by combining data from the Watson strand of DNA and the Crick strand of the nearby region for detecting arbitrary base modifications according to embodiments of the present invention. FIG. 10 shows an example of a Watson strand, a Crick strand, and a measurement window for both strands for determining the methylation status of a CpG site, according to embodiments of the present invention. FIG. 1 shows general procedures for building analytical, computational, mathematical, or statistical models for classifying base modifications according to embodiments of the invention. 1 shows a general procedure for classifying base modifications according to embodiments of the invention. General building analytical, computational, mathematical, or statistical models for classifying the methylation status of CpG sites using samples with known methylation status of Watson strands according to embodiments of the present invention step-by-step instructions. 1 shows a general procedure for classifying the Watson strand methylation status of unknown samples according to embodiments of the present invention. Building analytical, computational, mathematical, or statistical models for classifying methylation states at CpG sites using samples with known methylation states of click strands according to embodiments of the present invention Here are the general steps. 1 shows a general procedure for classifying the click strand methylation status of an unknown sample according to embodiments of the present invention. A general procedure for building a statistical model for classifying the methylation status of CpG sites using samples with known methylation status from both Watson and Crick strands according to embodiments of the present invention. show. FIG. 2 shows a general procedure for classifying the methylation status of unknown samples from Watson and Crick strands according to embodiments of the present invention. FIG. Figure 3 shows the performance of training and test datasets for determining methylation according to embodiments of the invention. Ditto. Figure 3 shows the performance of training and test datasets for determining methylation according to embodiments of the invention. Ditto. FIG. 4 shows the performance of training and test datasets at different sequencing depths for determining methylation according to embodiments of the invention. FIG. Ditto. FIG. 4 shows the performance of different strand training and test datasets for determining methylation according to embodiments of the present invention. FIG. Ditto. FIG. 4 shows the performance of training and test datasets for different measurement windows for determining methylation according to embodiments of the invention. FIG. Ditto. FIG. 4 shows the performance of training and test datasets for different measurement windows using downstream bases only to determine methylation, according to embodiments of the present invention. FIG. Ditto. FIG. 4 shows the performance of training and test datasets for different measurement windows using upstream bases only to determine methylation, according to embodiments of the present invention. FIG. Ditto. Figure 2 shows the performance of methylation analysis using dynamic patterns associated with downstream and upstream bases using asymmetric neighborhood sizes in the training dataset, according to embodiments of the present invention. FIG. 4 shows the performance of methylation analysis using dynamic patterns associated with downstream and upstream bases using asymmetric neighborhood sizes in test datasets, according to embodiments of the present invention. Fig. 2 shows the relative importance of features for classifying the methylation status of CpG sites according to embodiments of the present invention. Figure 2 shows the performance of motif-based IPD analysis for methylation detection without the use of pulse width signals, according to embodiments of the present invention. FIG. 10 is a graph of a principal component analysis technique using 2 nts upstream and 6 nts downstream of cytosine subjected to methylation analysis according to embodiments of the present invention; FIG. 5 is a graph of a performance comparison between a method using principal component analysis and a method using convolutional neural networks, according to embodiments of the invention; FIG. 4 shows training and test data set performance of different analytical, computational, mathematical, or statistical models using upstream bases only to determine methylation, according to embodiments of the present invention. Ditto. An example of one approach for generating molecules with unmethylated adenines by whole genome amplification according to embodiments of the present invention is shown. An example of one approach for generating molecules with methylated adenines by whole genome amplification according to embodiments of the present invention is shown. FIG. 4 shows pulse interval (IPD) values across sequenced A bases in Watson strand template DNA between unmethylated and methylated data sets, according to embodiments of the present invention. Ditto. FIG. 4 shows a receiver operating characteristic curve for determining Watson strand methylation, according to embodiments of the present invention. FIG. FIG. 4 shows pulse interval (IPD) values across sequenced A bases in click strand template DNA between unmethylated and methylated data sets, according to embodiments of the present invention. Ditto. FIG. 4 shows receiver operating characteristic curves for determining click strand methylation, according to embodiments of the present invention. FIG. 6 shows the Watson chain 6 mA determination according to embodiments of the present invention. 6 shows a click strand 6mA determination according to an embodiment of the present invention. Determined probabilities of being methylated for sequenced A bases of the Watson strand between the uA and mA datasets using a measurement window-based convolutional neural network model, according to embodiments of the present invention. indicates Ditto. FIG. 10 shows a ROC curve for detecting 6 mA using a measurement window-based CNN model of sequenced A bases of Watson strands according to embodiments of the present invention. FIG. 6 shows a performance comparison between IPD metric-based 6mA detection and measurement window-based 6mA detection according to embodiments of the present invention. Using a measurement window-based CNN model according to embodiments of the present invention, the determined probabilities of being methylated for those sequenced A bases in the click strand between the uA and mA data sets are show. Ditto. FIG. 4 shows the performance of 6 mA detection using a measurement window-based CNN model for sequenced A bases of click strands according to embodiments of the present invention. FIG. 4 shows examples of methylation states across the A bases of molecules comprising Watson and Crick strands, according to embodiments of the present invention. FIG. 11 shows an example of reinforcement training by selectively using A bases of the mA data set with IPD values above its 10th percentile, according to embodiments of the present invention. FIG. FIG. 10 is a graph of the percentage of unmethylated adenine in the mA data set versus the number of subreads in each well, according to embodiments of the present invention; FIG. 4 shows the pattern of methyladenines between the Watson and Crick strands of a double-stranded DNA molecule in a test data set, according to embodiments of the present invention. 4 is a table showing the percentage of fully unmethylated, hemimethylated, fully methylated and molecules with interlaced methyladenine patterns in training and test datasets, according to embodiments of the present invention. Representative examples of molecules with fully unmethylated, hemimethylated, fully methylated, and interlaced methyladenine patterns with respect to adenine sites are shown according to embodiments of the present invention. An example of a long read (6,265 bp) with CpG islands (yellow shading) according to embodiments of the invention is shown. FIG. 4 is a table showing that nine DNA molecules were sequenced by Pacific Biosciences SMRT sequencing and that they overlap with imprinted regions, according to embodiments of the present invention. FIG. 1 shows an example of genomic imprinting according to embodiments of the present invention. FIG. 4 shows an example of determination of methylation patterns in imprinted regions according to embodiments of the present invention. FIG. Figure 2 shows a comparison of estimated methylation levels between the new approach and conventional bisulfite sequencing according to embodiments of the present invention. Figure 2 shows the performance of plasma DNA methylation detection according to embodiments of the present invention. (A) Relationship between the predicted probability of methylation and the range of methylation levels quantified by bisulfite sequencing. (B) Methylation levels determined by Pacific Biosciences (PacBio) sequencing (y-axis) and methylation levels quantified by bisulfite sequencing at 10 Mb resolution (x-axis) according to embodiments present in the present disclosure. ). Ditto. FIG. 2 shows the correlation of Y chromosome genomic representation (GR) between Pacific Biosciences SMRT sequencing and BS-seq, according to embodiments of the present invention. FIG. 4 shows an example of CpG block-based detection of methylation using CpG blocks each having a series of CpG sites according to embodiments of the invention. 5mC: methylated, C: unmethylated. Figure 3 shows training and testing of methylation calling of human DNA molecules using a CpG block-based approach according to embodiments of the present invention. (A) Performance of the training dataset. (B) Performance of independent test datasets. Ditto. Figure 3 shows copy number alterations in tumor tissue according to embodiments of the present invention. Ditto. Figure 3 shows copy number alterations in tumor tissue according to embodiments of the present invention. Ditto. FIG. 4 shows a schematic of plasma DNA tissue mapping from plasma of pregnant women using estimated methylation levels according to embodiments of the present invention. FIG. 4 shows a correlation between the estimated placental contribution to maternal plasma DNA and the fetal DNA fraction estimated by Y-chromosome reads, according to embodiments of the present invention. 1 shows a table summarizing sequencing data from different human tissue DNA samples, according to embodiments of the present invention. FIG. 2 shows diagrams of various methods of analyzing methylation patterns, according to embodiments of the present invention. FIG. 4 shows a comparison of methylation densities at the genome-wide level as quantified by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. Ditto. Figure 3 shows different correlations of global methylation levels quantified by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. Ditto. Ditto. Methylation levels for hepatocellular carcinoma (HCC) cell lines and buffy coat samples from healthy control subjects, along with methylation levels determined by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. Patterns are shown at 1 Mnt resolution. Ditto. Scattering methylation levels at 1 Mnt resolution determined by bisulfite sequencing and single-molecule real-time sequencing according to embodiments of the present invention for buffy coat samples from HCC cell lines (HepG2) and healthy control subjects. Figure shows. Ditto. Scattering methylation levels at 100 knt resolution determined by bisulfite sequencing and single molecule real-time sequencing according to embodiments of the present invention for buffy coat samples from HCC cell lines (HepG2) and healthy control subjects. Figure shows. Ditto. Methylation patterns for HCC tumor tissue and adjacent normal tissue are shown at 1 Mnt resolution, along with methylation levels determined by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. Ditto. Figure 3 shows a scatter plot of methylation levels at 1 Mnt resolution determined by bisulfite sequencing and single molecule real-time sequencing according to embodiments of the present invention for HCC tumor tissue and adjacent normal tissue. Ditto. Figure 3 shows a scatter plot of methylation levels at 100 knt resolution determined by bisulfite sequencing and single molecule real-time sequencing according to embodiments of the present invention for HCC tumor tissue and adjacent normal tissue. Ditto. Methylation patterns for HCC tumor tissue and adjacent normal tissue are shown at 1 Mnt resolution, along with methylation levels determined by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. Ditto. Figure 3 shows a scatter plot of methylation levels at 1 Mnt resolution determined by bisulfite sequencing and single molecule real-time sequencing according to embodiments of the present invention for HCC tumor tissue and adjacent normal tissue. Ditto. Figure 3 shows a scatter plot of methylation levels at 100 knt resolution determined by bisulfite sequencing and single molecule real-time sequencing according to embodiments of the present invention for HCC tumor tissue and adjacent normal tissue. Ditto. FIG. 11 shows an example of an aberrant pattern of methylation near the tumor suppressor gene CDKN2A according to embodiments of the present invention. FIG. FIG. 4 shows variable methylated regions detected by single-molecule real-time sequencing, according to embodiments of the present invention. Ditto. FIG. 4 shows hepatitis B virus DNA methylation patterns between HCC tissue and adjacent non-tumor tissue using single-molecule real-time sequencing, according to embodiments of the present invention. FIG. 4 shows hepatitis B virus DNA methylation levels in liver tissue from patients with cirrhosis but without HCC using bisulfite sequencing, according to embodiments of the present invention. FIG. 4 shows hepatitis B virus DNA methylation levels in HCC tissue using bisulfite sequencing, according to embodiments of the present invention. 4 shows methylation haplotype analysis according to embodiments of the invention. Figure 3 shows the size distribution of sequenced molecules determined from consensus sequences, according to embodiments of the present invention. FIG. 4 shows examples of allelic methylation patterns in imprinted regions, according to embodiments of the present invention. FIG. Ditto. Ditto. Ditto. FIG. 4 shows examples of allelic methylation patterns in non-imprinted regions according to embodiments of the present invention. FIG. Ditto. Ditto. Ditto. FIG. 4 shows a table of methylation levels of allele-specific fragments according to embodiments of the invention. FIG. FIG. 10 shows an example of using methylation profiles to determine the placental origin of plasma DNA during pregnancy, according to embodiments of the present invention. FIG. Figure 3 shows an analysis of fetal-specific DNA methylation according to embodiments of the present invention; Figure 2 shows the performance of different measurement window sizes across different reagent kits for SMRT-seq according to embodiments of the present invention. Ditto. Ditto. Figure 2 shows the performance of different measurement window sizes across different reagent kits for SMRT-seq according to embodiments of the present invention. Ditto. Ditto. Figure 2 shows the correlation of global methylation levels quantified by bisulfite sequencing and SMRT-seq (Sequel II Sequencing Kit 2.0) according to embodiments of the present invention. Ditto. Ditto. FIG. 4 shows a comparison of global methylation levels between various tumor tissues and paired adjacent non-tumor tissues, according to embodiments of the present invention. Ditto. FIG. 11 shows the use of sequence context determined from a circular consensus sequence (CCS) to determine methylation status, according to embodiments of the present invention. FIG. FIG. 4 shows ROC curves for detection of methylated CpG sites using sequence context determined from CCS, according to embodiments of the present invention. FIG. FIG. 4 shows ROC curves for detection of methylated CpG sites without CCS information and without prior alignment to the reference genome, according to embodiments of the present invention. FIG. 1 shows an example of molecule preparation for single-molecule real-time sequencing, according to embodiments of the present invention. 1 shows a diagram of a CRISPR/Cas9 system, according to an embodiment of the invention; FIG. FIG. 11 shows an example of a Cas9 complex for introducing two truncations spanning an end-blocked molecule of interest, according to embodiments of the present invention. FIG. FIG. 4 shows the methylation distribution of Alu regions determined by bisulfite sequencing and single-molecule real-time sequencing, according to embodiments of the present invention. FIG. 4 shows the distribution of methylation levels of Alu regions as determined by a model using single-molecule real-time sequencing results, according to embodiments of the present invention. FIG. 3 shows a table of tissues and methylation levels of Alu regions in tissues according to embodiments of the present invention. FIG. FIG. 4 shows cluster analysis of different cancer types using methylation signals associated with Alu repeats according to embodiments of the present invention. FIG. Whole genome amplification and M . Figure 2 shows the effect of read depth on quantification of global methylation levels in test datasets involving SsssI treatment. Ditto. Figure 2 shows a comparison between global methylation levels determined by SMRT-seq (Sequel II Sequencing Kit 2.0) and BS-seq using different sub-read depth cutoffs according to embodiments of the present invention. FIG. 11 is a table showing the effect of sub-read depth on the correlation of methylation levels between two measurements by SMRT-seq (Sequel II Sequencing Kit 2.0) and BS-seq, according to embodiments of the present invention. FIG. FIG. 4 shows sub-read depth distributions with respect to fragment size in data generated by Sequel II Sequencing Kit 2.0, according to embodiments of the present invention. FIG. FIG. 1 illustrates a method for detecting nucleotide modifications of nucleic acid molecules according to embodiments of the present invention. FIG. FIG. 1 illustrates a method for detecting nucleotide modifications of nucleic acid molecules according to embodiments of the present invention. FIG. FIG. 3 shows a relative haplotype-based methylation imbalance analysis according to embodiments of the present invention. FIG. Table of haplotype blocks showing differential methylation levels between haplotype I (Hap I) and haplotype II (Hap II) in tumor DNA compared to adjacent non-tumor tissue DNA of case TBR3033, according to embodiments of the present invention. is. Ditto. FIG. 10 is a table of haplotype blocks showing differential methylation levels between Hap I and Hap II in tumor DNA compared to adjacent normal tissue DNA of case TBR3032, according to embodiments of the present invention. FIG. Summarize the number of haplotype blocks showing methylation imbalance between two haplotypes between tumor and adjacent non-tumor tissue based on data generated by Sequel II Sequencing Kit 2.0 according to embodiments of the present invention. It is a table. FIG. 11 is a table summarizing the number of haplotype blocks showing methylation imbalance between two haplotypes in tumor tissues of different tumor stages, based on data generated by Sequel II Sequencing Kit 2.0, according to an embodiment of the present invention; be. FIG. 3 shows a relative haplotype-based methylation imbalance analysis according to embodiments of the present invention. FIG. FIG. 4 illustrates a method of classifying disorders in organisms having a first haplotype and a second haplotype, according to embodiments of the present invention; FIG. Figure 3 shows the generation of a human-mouse hybrid fragment in which the human portion is methylated, but the mouse portion is unmethylated, according to embodiments of the present invention. Figure 3 shows the generation of a human-mouse hybrid fragment in which the human portion is unmethylated, but the mouse portion is methylated, according to embodiments of the present invention. Figure 2 shows the chain length distribution of DNA molecules in a DNA mixture (sample MIX01) after ligation according to an embodiment of the present invention. FIG. 2 shows junction regions where a first DNA (A) and a second DNA (B) are joined together according to embodiments of the present invention. Figure 3 shows methylation analysis of DNA mixtures according to embodiments of the present invention. FIG. 10 shows a boxplot of methylated probabilities for CpG sites in sample MIX01, according to embodiments of the present invention. FIG. Figure 2 shows the chain length distribution of DNA molecules in the DNA mixture after cross-ligation of sample MIX02 according to embodiments of the present invention. FIG. 10 shows a boxplot of methylated probabilities for CpG sites in sample MIX02, according to embodiments of the present invention. FIG. FIG. 4 is a table comparing methylation determined by bisulfite sequencing and Pacific Biosciences sequencing of MIX01, according to embodiments of the present invention. FIG. FIG. 4 is a table comparing methylation determined by bisulfite sequencing and Pacific Biosciences sequencing of MIX02, according to embodiments of the present invention. FIG. FIG. 4 shows methylation levels in 5 Mb bins of human-only DNA and mouse-only DNA for MIX01 and MIX02, according to embodiments of the present invention. Ditto. FIG. 4 shows methylation levels in 5 Mb bins of human and mouse portions of human-mouse hybrid DNA fragments for MIX01 and MIX02 according to embodiments of the present invention. Ditto. FIG. 4 is a representative graph showing methylation status in a single human-mouse hybrid molecule, according to embodiments of the present invention. FIG. Ditto. Figure 4 shows a method of detecting chimeric molecules in a biological sample according to embodiments of the present invention. 1 shows a measurement system according to an embodiment of the invention; 1 shows a block diagram of an exemplary computer system usable with the systems and methods according to embodiments of the present invention; FIG. Figure 3 shows MspI-based targeted single-molecule real-time sequencing using DNA end repair and A-tailing according to embodiments of the present invention. Figure 2 shows the size distribution of MspI digested fragments according to embodiments of the present invention. Ditto. FIG. 4 shows a table of numbers of DNA molecules for certain selected size ranges, according to embodiments of the present invention. FIG. 4 is a graph of percent coverage of CpG sites within a CpG island versus DNA fragment size after restriction enzyme digestion, according to embodiments of the present invention. Figure 3 shows MspI-based targeted single-molecule real-time sequencing without DNA end repair and A-tailing according to embodiments of the present invention. FIG. 11 shows MspI-based targeted single-molecule real-time sequencing with reduced probability of adapter self-ligation according to embodiments of the present invention. FIG. FIG. 4 is a graph of global methylation levels between placenta and buffy DNA samples determined by MspI-based targeted single-molecule real-time sequencing, according to embodiments of the present invention. FIG. Figure 3 shows cluster analysis of placenta and buffy coat samples using DNA methylation profiles determined by MspI-based targeted single-molecule real-time sequencing, according to embodiments of the present invention.

The term "tissue" corresponds to a group of cells grouped together as a functional unit. More than one type of cell can be found within a single tissue. Different types of tissue may be composed of different types of cells (e.g., hepatocytes, alveolar cells, or blood cells) and may be affected by different organisms (maternal versus fetal, transplanted target tissue, microbial or viral). tissue from an infected organism) or healthy versus tumor cells. A "reference tissue" corresponds to a tissue used to determine tissue-specific methylation levels. Multiple samples of the same tissue type from different individuals can be used to determine tissue-specific methylation levels for that tissue type.

"Biological sample" refers to any sample taken from a human subject. A biological sample can be a tissue biopsy, fine needle aspirate, or blood cells. The sample can also be, for example, plasma or serum or urine from pregnant women. A stool sample may also be used. In various embodiments, the majority of the DNA in a biological sample from a pregnant woman enriched for cell-free DNA (e.g., a plasma sample obtained via a centrifugation protocol) can be cell-free, e.g., 50% More than, 60%, 70%, 80%, 90%, 95%, or 99% of the DNA can be cell-free. A centrifugation protocol can include, for example, obtaining a fluid portion at 3,000 g×10 min and re-centrifuging at 30,000 g for an additional 10 min to remove residual cells. In certain embodiments, the 3,000 g centrifugation step can be followed by filtration of the fluid portion (eg, using filters with pore sizes of 5 μm diameter or less).

A "sequence read" refers to a strand of nucleotides that is sequenced from any portion or all of a nucleic acid molecule. For example, a sequence read can be a short nucleotide sequence (eg, about 20-150) from a nucleic acid fragment, short nucleotides at one or both ends of a nucleic acid fragment, or sequencing of an entire nucleic acid fragment present in a biological sample. can be Sequence reads may be obtained by polymerase chain reaction (PCR), e.g., using sequencing techniques or in various ways using probes, e.g., with hybridization arrays or capture probes, or using single primers or isothermal amplification. Alternatively, it can be obtained by an amplification technique such as linear amplification.

A "subread" is a sequence generated from all the bases of one strand of a circularized DNA template and copied into one contiguous strand by a DNA polymerase. For example, a subread can correspond to one strand of DNA in a circularized DNA template. In such an example, after circularization, one double-stranded DNA molecule has two subreads (one for each sequencing pass). In some embodiments, the generated sequence may contain a subset of all bases of one strand, eg, due to the presence of sequencing errors.

A "site" (also called a "genomic site") can be a single base position, or a group of correlated base positions, e.g., a CpG site, or a larger group of correlated base positions. handle. A "locus" can correspond to a region that includes multiple sites. A locus can contain only one site that would make the locus equivalent to the site in its context.

"Methylation state" refers to the state of methylation at a given site. For example, a site is either methylated, unmethylated, or possibly undetermined.

A "methylation index" for each genomic site (e.g., CpG site) indicates methylation at that site over the total number of reads covering that site (e.g., as determined from sequence reads or probes). It can refer to the percentage of DNA fragments. A "read" can correspond to information obtained from a DNA fragment (eg, the methylation state of a site). Reads can be obtained using reagents (eg, primers or probes) that preferentially hybridize to DNA fragments with specific methylation states at one or more sites. Typically, such reagents are involved in processes that differentially modify or recognize DNA molecules, depending on their methylation state, such as bisulfite conversion, or methylation-sensitive restriction enzymes, or methylation. treated with binding proteins, or anti-methylcytosine antibodies, or single-molecule sequencing technologies that recognize methylcytosine and hydroxymethylcytosine (e.g., single-molecule real-time sequencing and nanopore sequencing (e.g., from Oxford Nanopore Technologies)) applied later.

The "methylation density" of a region can refer to the number of reads at a site within the region, representing the methylation divided by the total number of reads covering the site in that region. This site may have specific characteristics and may be, for example, a CpG site. Thus, the "CpG methylation density" of a region is the CpG methylation divided by the total number of reads covering CpG sites in this region (e.g., a particular CpG site, CpG sites within a CpG island or larger region). Refers to the number of reads that indicate For example, the methylation density of each 100 kb bin in the human genome is expressed as the percentage of all CpG sites covered by sequence reads mapped to the 100 kb region, bisulfite treatment of CpG sites (corresponding to methylated cytosines) It can later be determined from the total number of unconverted cytosines. This analysis can also be performed for other bin sizes such as 500 bp, 5 kb, 10 kb, 50 kb, or 1 Mb. A region can be an entire genome, or a chromosome, or a portion of a chromosome (eg, a chromosomal arm). The methylation index of a CpG site is the same as the methylation density of the region if the region contains only that CpG site. "Percentage of methylated cytosines" is the total number of analyzed cytosine residues in this region, i.e., cytosines outside the context of the CpG that are methylated (e.g., unconverted after bisulfite conversion). The number of cytosine sites 'C' shown can be referred to. Examples of "methylation level" include methylation index, methylation density, number of molecules methylated at one or more sites, and number of molecules (e.g., cytosine) methylated at one or more sites. there is a proportion. Apart from bisulfite conversion, other processes known to those skilled in the art can be used to examine the methylation state of a DNA molecule, including but not limited to enzymes that are sensitive to the methylation state (e.g., methylation-sensitive restriction enzymes). ), methylation-binding proteins, single-molecule sequencing using platforms sensitive to methylation status (e.g., nanopore sequencing (Schreiber et al. Proc Natl Acad Sci 2013; 110: 18910-18915) and single-molecule real-time sequencing (eg by Pacific Biosciences) (Flusberg et al. Nat Methods 2010;7: 461-465)).

A "methylome" provides a measure of the amount of DNA methylation at multiple sites or loci in the genome. A methylome can correspond to the entire genome, a substantial portion of the genome, or relatively few point(s) of the genome.

A "pregnant plasma methylome" is a methylome determined from the plasma or serum of a pregnant animal (eg, human). The maternal plasma methylome is an example of a cell-free methylome because plasma and serum contain cell-free DNA. The maternal plasma methylome is also an example of a mixed methylome, as it is a mixture of DNA from different organs or tissues or cells in the body. In one embodiment, such cells include, but are not limited to, cells of erythroid (i.e., red cells) lineage, myeloid lineage (e.g., neutrophils and their precursors), and megakaryocytic lineage. is. During pregnancy, the plasma methylome may contain methylome information from the fetus and the mother. A "cellular methylome" corresponds to a methylome determined from a patient's cells (eg, blood cells). The methylome of blood cells is called a blood cell methylome (or blood methylome).

A "methylation profile" includes information relating to DNA or RNA methylation at multiple sites or regions. Information related to DNA methylation includes the methylation index of CpG sites, the methylation density (abbreviated MD) of CpG sites in a region, the distribution of CpG sites over contiguous regions, the distribution of CpG sites in regions containing two or more CpG sites. and non-CpG methylation at each individual CpG site. In one embodiment, a methylation profile may include patterns of methylation or unmethylation of more than one type of base (eg, cytosine or adenine). The methylation profile of a substantial portion of the genome can be considered equivalent to the methylome. "DNA methylation" in mammalian genomes typically refers to the addition of a methyl group to the 5' carbon of a cytosine residue (ie, 5-methylcytosine) between CpG dinucleotides. DNA methylation can occur at cytosines in other contexts, such as CHG and CHH, where H is adenine, cytosine, or thymine. Cytosine methylation can also be in the form of 5-hydroxymethylcytosine. Non-cytosine methylations such as N ⁶ -methyladenine have also been reported.

"Methylation pattern" refers to the order of methylated and unmethylated bases. For example, the methylation pattern can be the order of methylated bases on a single DNA strand, a single double-stranded DNA molecule, or another type of nucleic acid molecule. As an example, three consecutive CpG sites can have any of the following methylation patterns: UUU, MMM, UMM, UMU, UUM, MUM, MUU, or MMU. Here, "U" indicates an unmethylated site and "M" indicates a methylated site. When extending this concept to base modifications, including but not limited to methylation, we will use the term "modification pattern" to refer to the order of modified and unmodified bases. For example, a modification pattern can be the order of modified bases on a single DNA strand, a single double-stranded DNA molecule, or another type of nucleic acid molecule. As an example, three consecutive potentially modifiable sites can have any of the following modification patterns: UUU, MMM, UMM, UMU, UUM, MUM, MUU, or MMU. Here, "U" indicates an unmodified site and "M" indicates a modified site. An example of a base modification that is not based on methylation is an oxidative change such as 8-oxoguanine.

The terms "hypermethylation" and "hypomethylation" refer to the methylation density of a single DNA molecule as measured by the methylation level of that single molecule, e.g., the methylated bases or nucleotides within that molecule. divided by the total number of methylatable bases or nucleotides in the molecule. Hypermethylated molecules are molecules with a single molecule methylation level above a threshold and can be defined for each application. This threshold can be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%. Hypomethylated molecules are molecules with a single molecule methylation level below a threshold and can be defined and varied from application to application. This threshold can be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%.

The terms "hypermethylation" and "hypomethylation" may also refer to the methylation level of a population of DNA molecules as measured by the methylation level of multiple molecules of these molecules. A hypermethylated population of molecules is a population in which more than one molecule has a methylation level equal to or greater than a threshold, and can be defined and varied from application to application. This threshold can be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%. A hypomethylated population of molecules, which is a population in which more than one molecule has a methylation level below a threshold, can be defined and varied from application to application. This threshold can be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%. In one embodiment, the population of molecules can be aligned to one or more selected genomic regions. In one embodiment, the selected genomic region(s) may be associated with diseases such as cancer, genetic disorders, imprinting disorders, metabolic disorders, or neurological disorders. Selected genomic region(s) are 50 nucleotides (nt), 100nt, 200nt, 300nt, 500nt, 1000nt, 2knt, 5knt, 10knt, 20knt, 30knt, 40knt, 50knt, 60knt, 70knt, 80knt, 90knt, 100knt , 200 knt, 300 knt, 400 knt, 500 knt, or 1 Mnt.

The term "sequencing depth" refers to the number of times a locus is covered by sequence reads aligned to that locus. A locus can be as small as a nucleotide, or as large as a chromosomal arm, or as large as an entire genome. Sequencing depth is expressed as 50x, 100x, etc., where 'x' refers to the number of times the locus is covered by sequence reads. Sequencing depth can also be applied to multiple loci or the entire genome, where x can refer to the average number of times the locus or haploid genome or the entire genome is sequenced, respectively. Ultra-deep sequencing can refer to a sequencing depth of at least 100x.

As used herein, the term "classification" refers to any number(s) or other characteristic(s) associated with a particular property of a sample. For example, a "+" symbol (or the word "positive") can mean that the sample is classified as having deletions or amplifications. The classification can be binary (eg, positive or negative) or have more levels of classification (eg, a scale of 1-10 or 0-1).

The terms "cutoff" and "threshold" refer to a predetermined number used in some operation. For example, a cutoff size can refer to the size above which fragments are excluded. The threshold can be a value above or below what a particular classification requires. Any of these terms can be used in any of these contexts. A cutoff or threshold may be a "reference value" or may be derived from a reference value that represents a particular classification or distinguishes between two or more classifications. Such reference values can be determined in a variety of ways, as understood by those skilled in the art. For example, a metric can be determined for two different cohorts of subjects with different known classifications, with a reference value as representative of one classification (e.g., the mean), or a value between two clusters of the metric (e.g. , selected to obtain the desired sensitivity and specificity). As another example, the reference value can be determined based on statistical analysis or sample simulation.

The term "level of cancer" refers to whether cancer is present (i.e., present or absent), the stage of the cancer, the size of the tumor, whether there are metastases, the body's total tumor burden, the cancer's response to treatment, and/or other measures of cancer severity (eg, cancer recurrence). Cancer levels can be numbers or other indicia, such as symbols, letters, and colors. The level can be zero. Levels of cancer can also include premalignant or precancerous conditions (conditions). Cancer levels can be used in a variety of ways. For example, screening can check to see if cancer is present in a person who was not previously known to have cancer. Evaluation can examine a person who has been diagnosed with cancer, monitor cancer progression over time, study the effectiveness of therapy, or determine prognosis. In one embodiment, the prognosis can be expressed as the likelihood that the patient will die from the cancer, or the likelihood that the cancer will progress after a certain duration or time, or the likelihood or extent to which the cancer will metastasize. . Detecting can mean "screening," or it can mean checking to see if a person who has features (eg, symptoms or other positive tests) suggestive of cancer has cancer.

"Level of pathology" (or level of disorder) can refer to the amount, extent, severity of pathology associated with an organism, and can be as described above for cancer. Another example of pathology is rejection of transplanted organs. Examples of other pathologies include gene imprinting disorders, autoimmune attacks (e.g. lupus nephritis injury or multiple sclerosis that damages the kidney), inflammatory diseases (e.g. hepatitis), fibrotic processes (e.g. cirrhosis). , fatty infiltration (eg, fatty liver disease), degenerative processes (eg, Alzheimer's disease), and ischemic tissue damage (eg, myocardial infarction or stroke). A subject's healthy status can be considered a pathology-free classification.

A "pregnancy-related disorder" includes any disorder characterized by abnormal relative expression levels of genes in maternal and/or fetal tissues. These disorders include preeclampsia, intrauterine growth restriction, invasive placentation, preterm birth, hemolytic disease of the newborn, placental insufficiency, hydrops fetalis, fetal malformations, HELLP syndrome, systemic lupus erythematosus, and others. Includes but is not limited to maternal immune disorders.

The abbreviation "bp" refers to base pairs. In some cases, "bp" can be used to denote the length of a DNA fragment even when the DNA fragment is single-stranded and contains no base pairs. In the context of single-stranded DNA, "bp" may be taken to provide a length in nucleotides.

The abbreviation "nt" refers to nucleotide. In some cases, "nt" can be used to denote the length of a single-stranded DNA in bases. Also, "nt" can be used to denote a relative position, such as upstream or downstream of the analyzed locus. In some contexts relating to technical conceptualization, data representation, processing, and analysis, "nt" and "bp" may be used interchangeably.

The term "sequence context" can refer to the base composition (A, C, G, or T) and base order in a stretch of DNA. Such stretches of DNA may surround the bases to be subjected to base modification analysis or targeted bases. For example, sequence context can refer to bases upstream and/or downstream of the base subjected to base modification analysis.

The term "kinetic features" can refer to features derived from sequencing, including single-molecule real-time sequencing. Such features can be used for base modification analysis. Examples of kinetic features include upstream and downstream sequence context, strand information, pulse interval, pulse width, and pulse intensity. Single-molecule real-time sequencing continuously monitors the effect of polymerase activity on the DNA template. Measurements generated from such sequencing can therefore be viewed as kinetic features, eg, nucleotide sequences.

The term "machine learning model" may include models based on using sample data (eg, training data) to predict test data, and thus may include supervised learning. Machine learning models are often developed using computers or processors. Machine learning models can include statistical models.

The term "data analysis framework" can include algorithms and/or models that can take data as input and then output predictive results. Examples of "data analysis frameworks" include statistical models, mathematical models, machine learning models, other artificial intelligence models, and combinations thereof.

The term "real-time sequencing" may refer to techniques that involve data collection or monitoring while reactions involving sequencing are in progress. For example, real-time sequencing may involve optical monitoring or filming of DNA polymerases as they incorporate new bases.

The terms "about" or "approximately" can mean within a particular value tolerance range as determined by one skilled in the art, which depends in part on how the value is measured or determined, i.e., limitations of the measurement system. . For example, "about" can mean within 1 or more than 1 standard deviations, per the practice in the art. Alternatively, "about" can mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Alternatively, the term "about" or "approximately," particularly with respect to biological systems or processes, can mean within one order of magnitude, within five times, more preferably within two times the value. Where a particular value is recited in the present application and claims, the term "about" should be assumed within a tolerance range of the particular value unless otherwise stated. The term "about" may have the meaning commonly understood by those of ordinary skill in the art. The term "about" can refer to ±10%. The term "about" can refer to ±5%.

Achieving determination of bisulfite-free base modifications, including methylated bases, is the subject of various research efforts, none of which has been shown to be commercially viable. Recently, a bisulfite-free method for detecting 5mC and 5hmC was published using mild conditions for base conversion of 5mC and 5hmC (Y. Liu et al., 2019). The method involves multiple steps of enzymatic and chemical reactions, including ten eleven translocation (TET) oxidation, pyridine borane reduction, and PCR. The efficiency of each step of the conversion reaction as well as the PCR bias adversely affect the final accuracy of the 5mC analysis. For example, 5mC has been reported to have a conversion rate of about 96% and a false negative rate of about 3%. Such performance can limit the ability to detect certain subtle changes in methylation in the genome. Enzymatic conversions, on the other hand, may not work equally well across the genome. For example, conversion of 5hmC was 8.2% lower than that of 5mC, and conversion to non-CpG was 11.4% lower than that to CpG context (Y. Liu et al., 2019). . Therefore, the ideal situation would be an approach for measuring base modifications of natural DNA molecules without prior transformation (chemical or enzymatic, or a combination thereof) step and even without an amplification step. is to develop

There are several proof-of-concept studies (Q. Liu et al., 2019, Ni et al., 2019), using long-read nanopore sequencing approaches (e.g., systems developed by Oxford Nanopore Technologies). ) has allowed us to detect the methylation state using deep learning methods. In addition to the Oxford Nanopore, there are other single-molecule sequencing approaches that allow long reads. One example is single-molecule real-time sequencing. An example of single molecule real-time sequencing is commercialized as the Pacific Biosciences SMRT system. As a single-molecule principle, real-time sequencing (e.g., Pacific Biosciences SMRT system) is different from that of non-optical-based nanopore systems (e.g., Oxford Nanopore Technologies) for such non-optical-based nanopore systems. The developed base modification detection approach cannot be used for single molecule real-time sequencing. For example, non-optical nanopore systems have not been designed to capture the pattern of fluorescent signals generated by immobilized DNA polymerase-based DNA synthesis (employed in single-molecule real-time sequencing such as the Pacific Biosciences SMRT system). . As a further example, in the Oxford nanopore sequencing platform, each measured electrical event is associated with a k-mer (eg, 5-mer) (Q. Liu et al., 2019). However, on the Pacific Biosciences SMRT sequencing platform, each fluorescence event is generally associated with a single incorporated base. In addition, single DNA molecules are sequenced multiple times with Pacific Biosciences SMRT sequencing, including Watson and Crick strands. Conversely, for the Oxford Nanopore long-read sequencing approach, sequence reads are performed once for each of the Watson and Crick strands.

Polymerase kinetics have been reported to be affected by the methylation status of E. coli sequences (Flusberg et al., 2010). Previous studies have used single-molecule real-time sequencing to estimate the methylation state of specific CpGs in a single molecule (5mC versus C) when compared to the detection of 6mA, 4mC, 5hmC, and 8-oxoguanine. Using deterministic polymerase kinetics proved to be more difficult. This is because the methyl group is small, oriented in the major groove, does not participate in base-pairing, and provides very little disruption in the kinetics attributed to 5mC (Clark et al., 2013). Therefore, there is a paucity of approaches to determine the methylation status of cytosines at the single-molecule level.

Suzuki et al. developed an algorithm to combine the pulse interval (IPD) ratios of neighboring CpG sites in an attempt to increase confidence in identifying the methylation status of those sites (Suzuki et al., 2016). However, this algorithm was only able to predict genomic regions that were either fully methylated or not methylated at all, and lacked the ability to determine intermediate methylation patterns.

For single-molecule real-time sequencing, current approaches use only one or two parameters independently and have very high accuracy in detecting 5mC from the difference in measurements between 5-methylcytosine and cytosine. is limited to For example, Flusberg et al. demonstrated that the IPD changed at base modifications including N6-methyladenosine, 5-methylcytosine, and 5-hydroxymethylcytosine. However, no significant effect was found on the pulse width (PW) of sequencing kinetics. Therefore, the method they used to predict base modifications used detection of N6-methyladenosine, and only IPD, not PW, as an example.

In follow-up publications by the same group (Clark et al., 2012, Clark et al. 2013), IPD, rather than PW, was incorporated into the algorithm to detect 5-methylcytosine. In Clark et al. (2012), the detection rate of 5-methylcytosine that did not convert to 5-methylcytosine ranged from 1.9% to 4.3%. Furthermore, in Clark et al. (.2013) the authors further reconfirmed the subtleties of the kinetic signature of 5-methylcytosine. To overcome the poor detection sensitivity of 5-methylcytosine, Clark et al. A method was further developed to improve sensitivity to methylcytosine (Clark et al. 2013). This was because the change in IPD due to 5-carboxylcytosine was much greater than 5-methylcytosine.

A recent report by Blow et al. used the IPD ratio-based method previously described by Flusberg et al. detected (Blow et al., 2016). Of all the base modifications they identified, only 5% involved 5-methylcytosine. They attributed this low detection rate of 5-methylcytosine to the low sensitivity of single-molecule real-time sequencing for detecting 5-methylcytosine. In most bacteria, a series of sequence motifs are targeted for methylation by DNA methyltransferases (MTases) at nearly all of these motifs in the genome (e.g., 5'-GmATC-3' by Dam in E. coli). or 5′-CmCWGG-3′ by Dcm), only a small fraction of these motif sites remained unmethylated (Beaulaurier et al. 2019). In addition, an IPD-based method was used to classify the methylation status of the second C of the 5′-CCWGG-3′ motif, showing the detection rate of 5-methylcytosine with and without Tet protein treatment. were 95.2% and 1.9%, respectively (Clark et al. 2013). Overall, IPD methods without prior base conversion (eg, using the Tet protein) missed the majority of 5-methylcytosines.

In the studies mentioned above (Clark et al., 2012, Clark et al., 2013, Blow et al., 2016), IPD-based algorithms were used without considering the sequence context in which candidate base modifications are located. Other groups have attempted to consider the sequence context of nucleotides for the detection of base modifications. For example, Feng et al. analyzed IPD to detect 4-methylcytosine and 6-methyladenosine in their respective sequence contexts using a hierarchical model (Feng et al. 2013). However, their method only considered the IPD in the sequence context of the base of interest and its flanking bases, and did not use the IPD information of all neighboring bases flanking the base of interest. Furthermore, PW was not considered in the algorithm and no data on detection of 5-methylcytosine were presented.

In another study, Schadt et al. developed a statistical method called conditional random fields to analyze the IPD information of the base of interest and neighboring bases to determine whether the base of interest is 5-methylcytosine. (Schadt et al., 2012). In this study, the IPD interactions between those bases were considered by entering them into the equation. However, they did not enter the nucleotide sequence, ie A, T, G, or C, into their equations. When they applied this method, M. When the methylation status of the Sau3AI plasmid was determined, the area under the ROC curve was close to 0.5 even at 800-fold sequence coverage of the plasmid sequence. Furthermore, in their method they did not consider PW in the analysis.

Yet another study by Beckman et al. compared the IPDs of all sequences sharing the same 4-nt or 6-nt motif in the genome between the target bacterial genome and the fully unmethylated genome (e.g., obtained through whole-genome amplification). ) (Beckman et al. 2014). The purpose of such analyzes was only to identify motifs more frequently affected by base modifications. In this study, they only considered the IPDs of potentially modified bases, but not the IPDs of neighboring bases or PWs. Their method was not informative for the methylation status of individual nucleotides.

In summary, these previous attempts, either utilizing IPD alone or utilizing sequence information of neighboring nucleotides in combination with IPD for grouping data, have been 5-5 with no meaningful or practical accuracy. It was not possible to determine the base modification of methylcytosine. In a recent review by Gouil et al., the authors concluded that detection of 5-methylcytosine in single molecules using single-molecule real-time sequencing is imprecise due to low signal-to-noise ratios (Gouil et al. et al., 2019). In these previous studies, it remains unclear whether it is feasible to use dynamic features for whole-genome methylomic analysis, especially complex genomes such as human, cancer and fetal genomes.

In contrast to previous studies, some embodiments of the methods described in this disclosure are based on measuring and utilizing IPD, PW, and sequence context for all bases within the measurement window. there is We believe that single base We thought that accurate measurement of base modifications (eg, mC detection) could be achieved with a resolution of . Sequence context refers to the base composition (A, C, G, or T) and the order of bases in a stretch of DNA. Such stretches of DNA may surround the bases to be subjected to base modification analysis or targeted bases. In one embodiment, the stretch of DNA may be proximal to the bases subjected to base modification analysis. In another embodiment, the stretch of DNA may be far away from the bases subjected to base modification analysis. A stretch of DNA can be upstream and/or downstream of the bases subjected to base modification analysis.

In one embodiment, the upstream and downstream sequence context, strand information, IPD, pulse width, and pulse intensity features used for base modification analysis are referred to as kinetic features.

Embodiments present in the present disclosure include, but are not limited to, cell lines, samples from organisms (e.g., solid organs, solid tissues, samples obtained via endoscopy, blood, or maternal plasma or serum or urine, chorionic villus biopsy, etc.), samples obtained from the environment (eg, bacteria, cellular contaminants), DNA obtained from food (eg, meat). In some embodiments, the methods presented in the present disclosure also use, for example, hybridization probes (Albert et al., 2007; Okou et al., 2007; Lee et al., 2011) or physical separation (such as size after a step in which a portion of the genome is first enriched using an approach based on the can be applied. Enzymatic or chemical transformations are not required for the invention to function, but in certain embodiments such transformation steps may be included to further enhance the performance of the invention.

Embodiments of the present disclosure allow for improved accuracy or practicality or convenience in detecting base modifications or measuring modification levels. Modifications can be detected directly. Embodiments can avoid enzymatic or chemical transformations that may not retain all modification information for detection. Additionally, certain enzymatic or chemical transformations may be incompatible with certain types of modifications. Embodiments of the present disclosure may also avoid amplification by PCR, which may not convey base modification information to the PCR product. Additionally, both strands of DNA can be sequenced together, thereby allowing the pairing of sequences from one strand with sequences complementary to the other strand. In contrast, pairing of such sequences is difficult because PCR amplification splits the two strands of double-stranded DNA.

The determined methylation profile, with or without enzymatic or chemical conversion, can be used for analysis of biological samples. In one embodiment, methylation profiles can be used to detect the origin of cellular DNA (eg, maternal or fetal, tissue, viral, or tumor). Detection of aberrant methylation profiles in tissues helps identify developmental disorders in individuals, as well as identifying and predicting tumors and malignancies. Imbalance in methylation levels between haplotypes can be used to detect disorders including cancer. Single-molecule methylation patterns can occur in chimeric DNA (e.g., between viruses and humans) and hybrid DNA (e.g., between two genes that are not normally fused in the native genome) or between two species (e.g., due to genetic or genomic engineering). can be specified.

Methylation analysis can be improved by reinforcement training, which involves narrowing down the data used in the training set. Certain regions may be targeted for analysis. In embodiments, such targeting may include enzymes capable of cleaving a DNA sequence or genome based on its sequence, alone or in combination with other reagent(s). In some embodiments, the enzyme is a restriction enzyme that recognizes and cuts a specific DNA sequence(s). In other embodiments, two or more restriction enzymes with different recognition sequences can be used in combination. In some embodiments, the restriction enzyme may or may not cut based on the methylation state of the recognition sequence. In some embodiments, the enzyme is an enzyme within the CRISPR/Cas family. For example, the genomic region of interest may be a CRISPR/Cas9 system or other guide RNA-based system (i.e., a short RNA sequence that binds to a complementary target DNA sequence and in the process directs the enzyme to the target genomic location to act). ) can be used to target. In some cases, methylation analysis may be possible without alignment to the reference genome.

I. Methylation Detection by Single-Molecular Real-Time Sequencing Embodiments of the present disclosure allow direct detection of base modifications without enzymatic or chemical transformations. Kinetic features (e.g., sequence context, IPD, PW) obtained through single-molecule real-time sequencing can be analyzed with machine learning to develop models that detect modifications or detect the absence of modifications. The level of modification can be used to determine the origin of the DNA molecule or the presence or level of disorder.

For illustration, using Pacific Biosciences SMRT sequencing as an example of single-molecule real-time sequencing, DNA polymerase molecules are placed at the bottom of wells that act as zero-mode waveguides (ZMWs). do. A ZMW is a nanophotonic device for confining light to a small viewing volume. This is a very small diameter hole that does not allow the propagation of light in the wavelength range used for detection, only the low emission of optical signal from dye-labeled nucleotides incorporated by the immobilized polymerase. It is detectable against a constant background signal (Eid et al., 2009). DNA polymerases catalyze the incorporation of fluorescently labeled nucleotides into complementary nucleic acid strands.

FIG. 1 shows examples of molecules with base modifications sequenced by single-molecule circular consensus sequencing. Molecules 102, 104 and 106 have base modifications. A DNA molecule (eg, molecule 106 ) can be ligated with a hairpin adapter to form ligated molecule 108 . Linked molecules 108 can then form circularized molecules 110 . Circularized molecules can bind to the immobilized DNA polymerase and initiate DNA synthesis. Molecules without base modifications can also be sequenced.

FIG. 2 shows an example of molecules with methylated and/or unmethylated CpG sites sequenced by single-molecule real-time sequencing. First, a DNA molecule will be ligated to the hairpin adapter to form a circularized molecule that will bind the immobilized DNA polymerase and initiate DNA synthesis. In FIG. 2, DNA molecule 202 is ligated with a hairpin adapter to form ligated molecule 204 . Linked molecules 204 then form circularized molecules 206 . Molecules without CpG sites can also be sequenced. Circular molecule 206 contains unmethylated CpG sites 208, which can still be sequenced.

Upon initiation of DNA synthesis, fluorochrome-labeled nucleotides are incorporated into newly synthesized strands by an immobilized polymerase based on a circular DNA template, leading to the emission of a light signal. Because the DNA template is circularized, the entire circular DNA template is passed through the polymerase multiple times (ie, one nucleotide of the DNA template is sequenced multiple times). Sequences generated from the process of passing all bases of a circularized DNA template completely through a DNA polymerase are called subreads. A single molecule within a ZMW will generate multiple subreads, as the polymerase can continue over the circular DNA template multiple times. In one embodiment, a subread may contain only a subset of the sequence, base modifications or other molecular information of the circular DNA template, in one embodiment due to the presence of sequencing errors.

As shown in FIG. 3, the arrival times and durations of the resulting fluorescence pulses will allow the polymerase kinetics to be measured. Pulse interval (IPD) is a metric for the length of time between two emission pulses, each of which would suggest fluorescently labeled nucleotides incorporated into nascent strands (FIG. 3). As shown in FIG. 3, pulse width (PW) is another metric that reflects polymerase dynamics in relation to the duration of pulses associated with base calls. PW can be the pulse duration (ie, the fluorescence intensity of the incorporated dye-labeled nucleotide) at 0% of the signal peak height. In one embodiment, PW is for example, but not limited to, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80% or 90% of the signal peak height. can be defined by the pulse duration of . In some embodiments, PW may be the area under the peak divided by the height of the signal peak.

Such polymerase kinetics, such as IPD, have been observed in synthetic and microbial sequences (e.g., E. coli) for bases such as N6-methyladenine (6mA), 5-methylcytosine (5mC), and 5-hydroxymethylcytosine (5hmC). It has been shown to be affected by modification (Flusberg et al., 2010). Flusberg et al. (.2010) did not use sequence context and IPD as independent inputs to detect modifications, resulting in a model that lacked substantially meaningful detection accuracy. Flusberg et al. used sequence context only to confirm that 6 mA occurred at GATC. Flusberg et al. do not mention using sequence context in combination with IPD as input for detecting methylation status.

The weak interruption afforded to the incorporation of new bases into 5-methylcytosines of the complementary strand has been reported to range from only 1.9% to 4.3% of detection of the methylation motif C ^m CWGG. (Clark et al., 2013), making determination of methylation very difficult even in relatively simple microbial genomes when using only the IPD signal. For example, the analysis software package provided by Pacific Biosciences (SMRT Link v6.0.0) cannot perform 5mC analysis. Additionally, previous versions of SMRT Link v5.1.0 required the conversion of 5mC to 5-carboxylcytosine (5caC) using the Tet1 enzyme prior to methylation analysis. This is due to the enhanced IPD signal associated with 5caC (Clark et al., 2013). It is therefore not surprising that there are no studies demonstrating the feasibility of analyzing natural DNA in a genome-wide fashion of the human genome using single-molecule real-time sequencing.

II. Measurement Window Patterns and Machine Learning Models Techniques to detect modifications of bases without enzymatic or chemical conversion of the modifications and/or bases are desired. As described herein, modifications of a target base can be detected using kinetic characterization data obtained from single-molecule real-time sequencing of bases surrounding the target base. Kinetic features can include pulse intervals, pulse widths, and sequence context. These kinetic features can be obtained for measurement windows of a specified number of nucleotides upstream and downstream of the target base. These features (eg, specific locations in the measurement window) can be used to train machine learning models. As an example of sample preparation, two strands of a DNA molecule can be joined by a hairpin adapter, thereby forming a circular DNA molecule. Circular DNA molecules allow one to obtain the kinetic characteristics of either or both Watson and Crick strands. A data analysis framework can be developed based on the dynamic characteristics of the measurement window. Modifications, including methylation, can then be detected using this data analysis framework. This section describes various techniques for detecting modifications.

A. Use of Single Strands As an example, Watson strand subreads were obtained from Pacific Biosciences SMRT sequencing to analyze one specific base for base modification status, as shown in FIG. In FIG. 4, three bases from each side of the base subjected to base modification analysis would be defined as measurement window 400 . In one embodiment, the sequence context, IPD, and PW for these 7 bases (i.e., 3 nucleotide (nt) upstream and downstream sequences and 1 nucleotide for base modification analysis) are measured in two dimensions ( 2-D) compiled into a matrix. In the example shown, the measurement window 400 is for one subread of the Watson strand. Other variations are described herein.

The first row 402 of the matrix shows the sequences investigated. In the second row 404 of the matrix, the 0 position represented the base for base modification analysis. Relative positions of -1, -2, and -3 indicated positions 1 nt, 2 nt, and 3 nt upstream of the base subjected to base modification analysis, respectively. Relative positions of +1, +2, and +3 indicated positions 1 nt, 2 nt, and 3 nt downstream of the base subjected to base modification analysis, respectively. Each position contains two columns containing the corresponding IPD and PW values. The next four rows (rows 408, 412, 416, and 420) corresponded respectively to the four nucleotides (A, C, G, and T) of the strand (eg Watson strand). The IPD and PW values present in the matrix depended on which corresponding nucleotide type was sequenced at a particular position. As shown in FIG. 4, at relative position 0, the IPD and PW values are displayed in the line indicating the "G" of the Watson chain, indicating that guanine was called for in the sequence results at that position. Other grids in columns that did not correspond to sequenced bases are coded as '0'. As an example, the sequence information corresponding to the 2D digital matrix (Figure 4) is 5'-GATGACT-3' for the Watson chain.

As shown in one embodiment illustrated in FIG. 5, a measurement window can be applied to the data from the click chain. To analyze one specific base for base modification status, click strand subreads were obtained from single-molecule real-time sequencing. In FIG. 5, three bases from each side of the base subjected to base modification analysis and the base subjected to base modification analysis would be defined as the measurement window. In one embodiment, the sequence context, IPD, PW for these 7 bases (i.e., 3 nucleotide (nt) upstream and downstream sequences and 1 nucleotide for base modification analysis) are two-dimensional (i.e. , 2-D) compiled into matrices. The first row of the matrix shows the sequences investigated. In the second row of the matrix, the 0 position represents the base of the base modification analysis. Relative positions of -1, -2, and -3 indicated positions 1 nt, 2 nt, and 3 nt upstream of the base subjected to base modification analysis, respectively. Relative positions of +1, +2, and +3 indicated positions 1 nt, 2 nt, and 3 nt downstream of the base subjected to base modification analysis, respectively. Each position contains two columns containing the corresponding IPD and PW values. The next four lines correspond to the four nucleotides (A, C, G, T) of this strand (eg, click strand). The IPD and PW values present in the matrix depended on which corresponding nucleotide type was sequenced at a particular position. As shown in FIG. 5, at relative position 0, the IPD and PW values are displayed in the line labeled 'T' in the click strand, indicating that thymine was called for in the sequence results at that position. Other grids in columns that did not correspond to sequenced bases are coded as '0'. As an example, the sequence information corresponding to the 2D digital matrix (Figure 5) is 5'-ACTTAGC-3' for the click strand.

B. Using Both Watson and Crick Strands FIG. 6 shows an embodiment in which the measurement window can be implemented in a way that data from the Watson strand and its complementary Crick strand can be combined. As shown in FIG. 6, Watson and Crick strand subreads were obtained from single-molecule real-time sequencing and analyzed for modification of one particular base. In one embodiment, the measurement window from the Crick strand of the circular DNA template was complementary to the measurement window from the Watson strand subjected to base modification analysis. In FIG. 6, three bases from each side of the first base and the first base of the Watson strand subjected to base modification analysis would be defined as the first measurement window. Three bases from each side of the second base of the click strand and the second base will be defined as the second measurement window. The second base was complementary to the first base. In one embodiment, the sequence context, IPD, PW, for these 7 bases from the Watson and Crick strand (i.e., 3 nucleotide (nt) upstream and downstream sequences and 1 nucleotide for base modification analysis) is 2 Compiled into a dimensional (ie, 2-D) matrix. These measurement windows from the Watson and Crick strands were considered the first and second measurement windows, respectively.

The first row of the matrices for Watson and Crick strands shows the sequences investigated. In the second row of the Watson strand matrix, the 0 position represents the first base of the base modification analysis. The 0 position shown in the second row of the click strand matrix represents the second base complementary to the first base. Relative positions of -1, -2, and -3 indicated positions 1 nt, 2 nt, and 3 nt upstream of the first and second bases, respectively. Relative positions of +1, +2, and +3 indicated positions 1 nt, 2 nt, and 3 nt downstream of the first and second bases, respectively. Each position from the Watson and Crick strands will correspond to two columns containing the corresponding IPD and PW values. The next four rows of matrices for Watson and Crick strands each corresponded to the four nucleotides (A, C, G, and T) of a particular strand (eg, Crick strand). The IPD and PW values present in the matrix depended on which corresponding nucleotide type was sequenced at a particular position.

As shown in FIG. 6, at relative position 0, the IPD and PW values are shown in the row labeled "A" for Watson strands and "T" for Crick strands, and for Watson and Crick strands at that position. Sequence results show that adenine and thymine were called, respectively. Other grids in columns that did not correspond to sequenced bases are coded as '0'. As an example, the sequence information corresponding to the Watson chain 2D digital matrix (FIG. 6) would be 5'-ATAAGTT-3'. The sequence information corresponding to the 2D digital matrix of the click strand (Figure 6) would be 5'-AACTTAT-3'.

As shown in this example, data from Watson and Crick strands can be combined to form a new matrix, which can also be viewed as a measurement window. This new matrix can be used as a single sample used to train a machine learning model. Therefore, although the particular arrangement of the 2D matrix may have implications, such as when a convolutional neural network (CNN) is used, all values of the new matrix can be treated as separate features. Sequence context at various positions on different strands can be conveyed via non-zero entries in the matrix.

FIG. 7 shows that the measurement window can be implemented in such a way that the data from the Watson and Crick strands are not exactly complementary to each other. As shown in Figure 7, the first measurement window was 5'-ATAAGTT-3' and the second measurement window was 5'-GTAACGC-3'. In some embodiments, the Watson and Crick strands may be shifted relative to each other such that they are not complementary in position.

FIG. 8 shows that measurement windows can be used to analyze the methylation status of CpG sites. The 0 position corresponds to a cytosine in the CpG site, thus shifting by one position between the two strands, resulting in C being the 0 position for both strands. Therefore, only part of the sequences included in the measurement windows from the Watson and Crick strands are complementary to each other. In other embodiments, all sequences of the measurement windows from the Watson and Crick strands can be complementary to each other. In still other embodiments, none of the measurement window sequences from the Watson strand and the Crick strand are complementary to each other.

In one embodiment, the length of the DNA stretch surrounding the bases subjected to base modification analysis can be asymmetric with respect to the measurement window. For example, the X-nt upstream and Y-nt downstream of that base can be used for base modification analysis. X is 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 , 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 , 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10,000. Y is 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 , 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 , 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10,000.

C. Model Training and Modification Detection FIG. 9 shows the general procedure for how to determine any base modification using measurement windows. Unmodified and known modified DNA samples were subjected to single-molecule real-time sequencing. A modified DNA (eg, modified molecule 902) means that a base (eg, base 904) has a modification (eg, methylation) at that site. Unmodified DNA (eg, unmodified molecule 906) means that the base (eg, base 908) has no modifications at that site. Both sets of DNA can be engineered or manipulated to form modified/unmodified DNA.

At stage 910, the sample can then undergo single-molecule real-time sequencing. As part of SMRT sequencing, circular molecules can be sequenced multiple times by repeated passages of immobilized DNA polymerase. Sequence information obtained each time is regarded as a sub-read. This allows one circular DNA template to generate multiple subreads. Sequencing subreads can be aligned to the reference genome using, for example, without limitation, BLASR (Mark J Chaisson et al, BMC Bioinformatics. 2012; 13: 238). In various other embodiments, BLAST (Altschul SF et al, J Mol Biol. 1990;215(3):403-410), BLAT (Kent WJ, Genome Res. 2002;12(4):656-664). , BWA (Li H et al, Bioinformatics. 2010; 26(5):589-595), NGMLR (Sedlazeck FJ et al, Nat Methods. 2018; 15(6):461-468), LAST (Kielbasa SM et al 2011; 21(3):487-493) and Minimap2 (Li H, Bioinformatics. 2018; 34(18):3094-3100) can be used to align subreads to the reference genome. . Alignment allows the identification of data for each subread at the same location so that data from multiple subreads can be combined (eg, averaged).

At stage 912, the IPD, PW, and sequence context surrounding the bases subjected to base modification analysis were obtained from the alignment results. At stage 914, the IPD, PW, and sequence context were recorded in a specific structure, such as, but not limited to, a 2D matrix as shown in FIG.

At stage 916, several 2D matrices containing molecules from reference kinetic patterns with known base modifications were used to train analytical, computational, mathematical, or statistical model(s). At stage 918, a statistical model resulting from training is developed. For simplicity, FIG. 9 shows only statistical models developed by training, but any model or data analysis framework can be developed. Examples of data analysis frameworks include machine learning models, statistical models, and mathematical models. Statistical models include linear regression, logistic regression, deep recurrent neural networks (e.g., long short-term memory, LSTM), Bayesian classifiers, hidden Markov models (HMM), linear discriminant analysis (LDA), k-means clustering, and noise. Concomitant applications include but are not limited to density-based spatial clustering (DBSCAN), random forest algorithms, and support vector machines (SVM). The stretch of DNA surrounding a base subjected to base modification analysis can be the X-nt upstream and Y-nt downstream of that base, the "measurement window."

The data structure can be used in the training process because the correct outputs (ie, modification states) are known. For example, IPD, PW, and sequence context corresponding to 3 nt upstream and downstream of bases from Watson strand and/or Crick strand(s) to train statistical model(s) to classify base modifications. can be used to construct the 2D matrix used in In this way, training can provide a model that can classify base modifications at nucleic acid positions that have previously known states.

FIG. 10 shows a general procedure for how statistical model(s) learned from DNA samples with known states of base modifications can detect base modifications. Samples with unknown base modification status were subjected to SMRT sequencing. Sequencing subreads were aligned to the reference genome using, for example, the techniques described above. Additionally or alternatively, the subreads can be aligned with each other. Still other embodiments may use only one subread or analyze them independently so that no alignment is performed.

For bases subjected to base modification analysis, IPD, PW , and the sequence context was obtained and associated with that base. In another embodiment, the measurement windows between training and testing procedures will be different. For example, the size of the measurement window between training and testing procedures may differ. These IPDs, PWs, and sequence contexts are transformed into 2D matrices. Such a 2D matrix of test samples would be compared to reference kinetic signatures to determine base modifications. For example, a 2D matrix of test samples can be compared to reference kinetic features through statistical model(s) learned from training samples, thus allowing base modifications at sites of nucleic acid molecules of test samples to be determined. Statistical models include linear regression, logistic regression, deep recurrent neural networks (e.g., long short-term memory, LSTM), Bayesian classifiers, hidden Markov models (HMM), linear discriminant analysis (LDA), k-means clustering, and noise. Concomitant applications include but are not limited to density-based spatial clustering (DBSCAN), random forest algorithms, and support vector machines (SVM).

FIG. 11 shows a general procedure for how methods can be developed for classifying the methylation status at CpG sites. DNA samples with known unmethylation and methylation at CpG sites were subjected to single-molecule real-time sequencing. Sequencing subreads were aligned to the reference genome. Watson chain data were used.

From the alignment results, the IPD, PW, and sequence context surrounding the cytosine at the CpG site subjected to methylation analysis was obtained and recorded in a 2D matrix, such as, but not limited to, the specific structure shown in FIG. rice field. Several 2D matrices containing reference kinetic patterns derived from molecules with known methylation status were used to train the statistical model(s). The stretch of DNA surrounding the base under investigation can be X-nt upstream and Y-nt downstream of that base, the "measurement window." X is 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 , 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 , 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10,000. Y is 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 , 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 , 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10,000. In one embodiment, the IPD, PW, and sequence context corresponding to the 3nt upstream and downstream of the base from the Watson strand are used to train a statistical model(s) for classifying base modifications. can be used to build

Figure 12 shows the general procedure for classifying the methylation status of unknown samples. Samples with unknown methylation status were subjected to single-molecule real-time sequencing. Sequencing subreads were aligned to the reference genome.

For the cytosines in the CG sites of the alignment results, the IPD, PW, and sequence context were obtained from the Watson strand using the equivalent measurement window applied in the training step (Fig. 11) to associate the modification with the base under investigation. Associated. These IPDs, PWs, and sequence contexts can be transformed into 2D matrices. Such a 2D matrix of test samples would be compared to the reference kinetic pattern shown in FIG. 11 to determine methylation status. X11

Figures 13 and 14 show that similar to the procedure with Watson strands, kinetic features from Crick strands can be used for training and testing procedures, as detailed above. The statistical model(s) can be the same model or different models. In the case of different models, they can be used to obtain independent classifications, which can be compared and, for example, if they are concordant, the modification state is identified. Then, if they do not match, an unclassified state can be identified. If they are of the same model, the data can be combined into a single data structure, eg the matrix of FIG.

Figures 15 and 16 show that kinetic features from both Watson and Crick strands can be used for training and testing procedures, as detailed above. DNA samples with known unmethylation and methylation at CpG sites were subjected to single-molecule real-time sequencing. Sequencing subreads were aligned to the reference genome, but it is also possible to align subreads to each other and can be done in other ways as described herein.

For the subreads of the alignment results, the IPD, PW, and sequence context surrounding the cytosines of the CpG sites subjected to methylation analysis were obtained. Because the DNA molecule has been circularized through the use of two hairpin adapters (e.g. following the SMRTBell template preparation protocol), the circular molecule can be sequenced more than once, thereby generating multiple subreads of the molecule. be. Subreads can be used to generate circular consensus sequence (CCS) reads. In general, for all methods described herein, one ZMW can generate multiple subreads, but only corresponds to one CCS read.

In some embodiments, the fully unmethylated dataset can be generated by PCR on human DNA fragments. For example, the full methylation data set is the CpG methyltransferase M. spp., where all CpG sites are assumed to be methylated. It can be generated via a human DNA fragment treated with SssI. In another example, M. Another CpG methyltransferase such as MpeI can be used. In other embodiments, restriction enzyme digestion and subsequent ligation of synthetic sequences with known methylation status or pre-existing DNA samples with different levels of methylation, or methylated and unmethylated DNA molecules (thus creating chimeras). The hybrid methylation state generated by the method (resulting in the proportion of methylated/unmethylated DNA molecules) can be used for training predictive models of methylation or classifiers.

Conversion of kinetic patterns, including sequence context, IPD, and pulse width (PW), into a 2D matrix containing features from Watson and Crick strands to analyze the methylation status of CG sites, as shown in FIG. can do. This approach allowed us to accurately capture the subtle dynamic changes due to methylated cytosines as well as the sequence context nearby. As with any of the various methods described herein, for each CpG present in the subread, measurement windows (e.g., 3 bases upstream and downstream of the cytosine at the CpG site) can be used for subsequent analysis, Therefore, a total of 7 nucleotides (including the cytosine of the CpG site) are analyzed together. The IPD and PW can be calculated for each base between those seven nucleotides. To capture the sequence context resulting from dynamic changes, the IPD and PW signals can be compiled into specific base calls, relative sequencing positions, and strand information, as shown in FIG. Such data structures are simply referred to as dynamic 2D digital matrices.

Such a 2D digital matrix is analogous to a "2D digital image". For example, the first row of the 2D digital matrix contained 3nt upstream and downstream of the cytosine site along with the relative positions surrounding the cytosine of the CpG locus subjected to methylation analysis. Position 0 represents the cytosine site at which methylation is to be determined. Relative positions of -1 and -2 indicated 1nt upstream and 2nt upstream of the cytosine in question. The +1 and +2 relative positions indicate 1 nt downstream and 2 nt downstream of the cytosine used. Each position will correspond to two columns containing the corresponding IPD and PW values. Each row corresponded to four nucleotides (A, C, G, and T) of the Watson and Crick strands. The entry of IPD and PW values in the matrix depends on the type of corresponding nucleotide preset to the sequence result (ie, subread) at a particular position.

As shown in Figure 15, at a relative position of 0, the IPD and PW values are shown in the "C" row of the Watson chain, suggesting that cytosine was called at that position. Other grids in columns that did not correspond to sequenced bases are coded as '0'. As an example, the sequence information corresponding to the 2D digital matrix (Figure 15) is 5'-ATACGTT-3' and 5'-TAACGTA-3' for Watson and Crick strands, respectively. In this context, the sequences flanking the cytosine upstream and downstream of the CpG sites of the Watson and Crick strands are different. Since methylation at CpG sites is symmetric between Watson and Crick strands (Lister et al., 2009), in one preferred embodiment, dynamics of both strands are used to generate a methylation prediction model trained. In another embodiment, Watson and Crick strands can be used separately to train a methylation prediction model.

Given the high data throughput of single-molecule real-time sequencing, in one embodiment, a deep learning algorithm (convolutional neural network (CNN)) (LeCun et al., 1989) distinguishes methylated from unmethylated CpGs. may be suitable for Other algorithms may additionally or alternatively be used such as, but not limited to, linear regression, logistic regression, deep recurrent neural networks (e.g., long-term short-term memory, LSTM), Bayesian classifiers, hidden Markov Models (HMM), linear discriminant analysis (LDA), k-means clustering, density-based spatial clustering for noisy applications (DBSCAN), random forest algorithms, support vector machines (SVM), and others. As described in FIGS. 6-8, training can use the Watson and Crick strands separately or in a new combined matrix.

Another transform of kinetic patterns can be an N-dimensional matrix. N can be 1, 3, 4, 5, 6, and 7, for example. For example, a 3D matrix is a stack of 2D matrices layered according to the number of tandem CG sites in the DNA stretch being analyzed, with the third dimension being the number of tandem CG sites in that DNA stretch. In some embodiments, the pulse intensity or pulse magnitude (eg, measured by the peak height of the pulse or by the area under the pulse signal) may also be incorporated into the matrix. The pulse intensity (pulse peak amplitude metric, FIG. 3) is added to an additional column adjacent to the columns associated with the PW and IPD values above the original 2D matrix, or added to the third dimension. A 3D matrix can be formed by either:

As a further example, a 2D matrix of 8 (rows) by 21 (columns) can be transformed into a 1D matrix (ie, vector) containing 168 elements. Also, this 1D matrix can be scanned to perform, for example, CNN and other modeling. As another example, the method can divide an 8x21 2D matrix into multiple smaller matrices, eg, two 4x21 2D matrices. Combining these two smaller matrices vertically yields a 3D matrix (ie, x=21, y=4, z=2). The method can scan a first 2D matrix and then a second 2D matrix to form a data representation for machine learning. The data can be further partitioned to form higher dimensional matrices. Additionally, secondary structure information can be added to the data structure, for example, additional matrices (1D matrices) can be added on top of the 2D matrices. Such additional matrices can encode whether each base within the measurement window participates in secondary structure (eg, stem-loop structure). For example, bases involved in a "stem" are coded as 0's and bases involved in a "loop" are coded as 1's.

In one embodiment, the methylation status of CpG sites within a single DNA molecule is methylated based on a statistical model rather than giving a qualitative result of "methylated" or "unmethylated" It can be expressed as a probability. A probability of 1 indicates that the CpG site can be considered methylated based on the statistical model. A probability of 0 indicates that the CpG site can be considered unmethylated based on the statistical model. In subsequent downstream analysis, the cutoff value can be used to classify whether a particular CpG site is classified as "methylated" or "unmethylated" based on probability. Possible values for cutoff include 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70% , 75%, 80%, 85%, 90%, or 95%. Those with a probability of being methylated for a CpG site greater than a given cutoff are classified as "methylated" and those with a probability of being methylated for a CpG site not greater than a given cutoff are classified as "non-methylated". categorized as A desired cutoff can be obtained from a training data set using, for example, receiver operating characteristic (ROC) curve analysis.

Figure 16 shows the general procedure for classifying the methylation status of unknown samples from Watson and Crick strands. Samples with unknown methylation status were subjected to single-molecule real-time sequencing. Sequencing subreads can be aligned to the reference genome or to each other, as well as other methods, to determine the consensus value (mean, median, mode, or other statistic) for a given position. As shown, the measurements for the two strands can be combined into a single 2D matrix.

For the cytosines of the CG sites in the alignment results, different size windows can be used, but equivalent measurement windows (CpG site 3 nts upstream and downstream of the cytosine) can be used to obtain the IPD, PW, and sequence context from the Watson strand. Such a 2D matrix of test samples can be compared to the reference kinetic pattern shown in FIG. 16 to determine methylation status.

III. Exemplary Model Training for Detecting Methylation To test the feasibility and validity of the proposed approach, prior to single-molecule real-time sequencing, M. et al. A placental DNA library was prepared using SssI treatment (methylated library) and PCR amplification (unmethylated library). 44,799,736 and 43,580,452 subreads of the methylated and unmethylated libraries were obtained, corresponding to 421,614 and 446,285 circular consensus sequences (CCS), respectively. As a result, each molecule was sequenced at a median of 34-fold and 32-fold in methylated and unmethylated libraries. The dataset was generated from DNA prepared by the Pacific Biosciences Sequel Sequencing Kit 3.0. This kit was developed for use with the original Pacific Biosciences Sequel sequencer. The original Sequel is referred to herein as Sequel I to distinguish it from its successor, Sequel II. Therefore, the Sequel Sequencing Kit 3.0 is referred to herein as the Sequel I Sequencing Kit 3.0. Sequencing kits designed for the Sequel II sequencer include Sequel II Sequencing Kit 1.0 and Sequel II Sequencing Kit 2.0, which are also described in this disclosure.

50% of the sequenced molecules generated from the methylated and unmethylated libraries were used to train the statistical model (the remaining 50% were used for validation). In this case, it is a convolutional neural network (CNN) model. As an example, a CNN model may have one or more convolutional layers (eg, 1D or 2D layers). A convolutional layer can use one or more different filters, each using a kernel that manipulates the local (e.g., nearby or surrounding) matrix values for a particular matrix element, which provides a new value for a particular matrix element. In one implementation, we used two 1D convolutional layers (each with 100 filters with a kernel size of 4). Filters can be applied individually and then combined (eg, in a weighted average). The resulting matrix can be smaller than the input matrix.

The convolutional layer is followed by a ReLU (Rectified Linear Unit) layer followed by a dropout layer with a dropout rate of 0.5. ReLU is an example of an activation function that manipulates individual values to obtain a new matrix (image) from the convolutional layer(s). Other activation functions (eg, sigmoid, softmax, etc.) can also be used. One or more of such layers can be used. The dropout layer can be used in the ReLU layer or the max pooling layer and acts as a regularizer to prevent overfitting. A dropout layer can be used during the training process to ignore different (e.g. random) values during the various iterations of the optimization process performed as part of the training (e.g. cost/loss function).

After the ReLU layer, a max pooling layer (eg pool size 2) can be used. A max pooling layer works similarly to a convolutional layer, but instead of taking the dot product between the input and the kernel, we can take the maximum of the regions from the input that overlap the kernel. Additional convolutional layer(s) can be used. For example, the data from the pooling layer can be input to another two 1D convolutional layers (eg, each with 128 filters with a kernel size of 2 followed by a ReLU layer), and the dropout rate A dropout layer of 0.5 can be used. A maximum pooling layer with a pool size of 2 was used. Finally, a fully connected layer (eg, with 10 neurons followed by a ReLU layer) can be used. An output layer with one neuron can be followed by a sigmoid layer, thus obtaining the probability of methylation. Various settings for layers, filters and kernel sizes can be adjusted. 468,596 and 432,761 CpG sites from methylated and unmethylated libraries were used in this training dataset.

A. Results for Training and Test Datasets FIG. 17A shows the probability of being methylated for each CpG site of each single DNA molecule in the training data set. The probability of methylation was much higher for the methylated library than for the unmethylated library. With a probability of being methylated cutoff of 0.5, 94.7% of unmethylated CpG sites were correctly predicted to be unmethylated, and 84.7% of methylated CpG sites were predicted to be methylated. correctly predicted.

FIG. 17B shows the performance of the test dataset. Models trained with the training dataset were used to predict the methylation status of 469,729 and 432,024 CpG sites in independent test datasets from methylated and unmethylated libraries. With a methylated probability cutoff of 0.5, 94.0% of unmethylated CpG sites were correctly predicted to be unmethylated, and 84.1% of methylated CpG sites were unmethylated. correctly predicted to exist. These results suggested that the use of novel transformations of kinetics in combination with sequence context may allow determination of the methylation status of DNA (eg, from human subjects).

By including a subset of features in the model, we evaluated the ability of each feature (sequence context, IPD, and PW) in predicting CpG methylation status. In the training data set, the (i) sequence context only, (ii) IPD only and (iii) PW only models had area under the curve (AUC) values of 0.5, 0.74 and 0.86, respectively. gave Combining IPD and sequence context improved performance with an AUC of 0.86. A combined analysis of sequence context (“Seq”), IPD, and PW significantly improved performance with an AUC of 0.94 (FIG. 18A). Performance of the independent test dataset was comparable to the training dataset (Fig. 18B).

The subread depth of a CpG site was defined as the average number of subreads covering the site and its surrounding 10 bp. As shown in FIGS. 19A and 19B, the higher the sub-read depth of the CpG sites, the higher the precision of methylation detection achieved. For example, as shown in the test data set (FIG. 19B), the AUC predictive of methylation status is 0.93 when each CpG site is at least 10 deep. However, with a sub-read depth of at least 300 for each CpG site, the AUC predicting methylation status is 0.98. On the other hand, even with a depth of 1, an AUC of 0.9 was achieved. This indicates that our approach achieves methylation prediction using low sequencing depth.

To test the effect of strand information on the performance of methylation analysis, sequence contexts, IPD and PW, derived from Watson and Crick strands, respectively, were used to train according to the embodiments present in this disclosure. Figures 20A and 20B show that for training and testing, a single strand, either Watson or Crick strand showed that it is feasible to use Using both strands, including the Watson and Crick strands (described for example in Figures 6-8), gave the best performance (AUC: 0.94 and 0.90 for training and test datasets, respectively). , suggesting that strand information is important to achieve optimal performance.

Different numbers of nucleotides upstream and downstream of the CpG site were further tested to study how this parameter affects performance according to the presently disclosed embodiments developed in this disclosure. Figures 21A and 21B show that the number of nucleotides upstream and downstream of a cytosine in the context of a CpG affects the accuracy of methylation prediction. For example, for exemplary purposes, consider but are not limited to 2 nucleotides (nt), 3nt, 4nt, 6nt, 8nt, 10nt, 15nt, and 20nt upstream and downstream of the cytosine under investigation, and 2nt of the cytosine under investigation. The AUC for the method using upstream and downstream is only 0.50 for both the training and test datasets, while the AUC for the method using 15 nt upstream and downstream of the investigated cytosine is 0.95. increases to 0.92. These results suggested that varying the length of the upstream and downstream regions flanking the cytosines analyzed would allow optimal performance to be found. In one embodiment, 3 nt upstream and downstream of cytosine can be used to determine methylation status and achieve an AUC of 0.89, as shown in FIG. 21B.

In one embodiment, asymmetric sequences flanking the investigated cytosine can be used to perform analysis according to embodiments present in the present disclosure. For example, 1 nt, 3 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt of cytosine In combination with downstream, 2 nt upstream can be used, 1 nt, 2 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt of cytosine, 3 nt upstream can be used in combination with 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt downstream; 4nt upstream can be used in combination with 13nt, 14nt, 15nt, 16nt, 17nt, 18nt, 19nt, 20nt, 25nt, 30nt, 35nt, and 40nt downstream. As another example, 1 nt, 3 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt of cytosine , and 2 nt downstream can be used in combination with 40 nt upstream and 1 nt, 2 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt of cytosine , 18nt, 19nt, 20nt, 25nt, 30nt, 35nt and 40nt upstream, 3nt downstream can be used in combination with cytosine 1nt, 2nt, 3nt, 5nt, 6nt, 7nt, 8nt, 9nt, 10nt, 11nt , 12nt, 13nt, 14nt, 15nt, 16nt, 17nt, 18nt, 19nt, 20nt, 25nt, 30nt, 35nt, and 40nt upstream can be used in combination with 4nt downstream. Provide improved accuracy in determining methylation status in certain embodiments by utilizing IPD, PW, chain information and sequence context associated with n-nt upstream and m-nt downstream of cytosine can do. Such different measurement windows can be applied to other types of base modification analysis, such as 5hmC, 6mA, 4mC, and oxoG, or any modification disclosed herein. Such different measurement windows can include DNA secondary structure analysis such as guanine quadruplex and stem-loop structures. Examples of such are described above. Such secondary structure information can also be added as another column of the matrix.

Figures 22A and 22B show that it is feasible to determine methylation status using kinetic patterns associated only with at least three downstream bases. According to embodiments present in the present disclosure, using features associated with cytosine and its downstream 3, 4, 6, 8, and 10 bases, in determining methylation status in a training dataset, AUC is: 0.91, 0.92, 0.94, 0.94, and 0.94, respectively; was 0.90.

However, Figures 23A and 23B show that the ability to discriminate methylation status appears to be diminished when using only features associated with upstream bases. In the training and test datasets, AUCs were all 0.50 for 2-10 upstream bases.

Figures 24 and 25 show that different combinations of upstream and downstream bases allow optimal classification to be achieved when determining methylation status. For example, features associated 8 bases upstream and 8 bases downstream of cytosine achieved the best performance in this dataset, with AUCs of 0.94 and 0.91 for the training and test datasets, respectively. rice field.

FIG. 26 shows the relative importance of features for classifying methylation status at CpG sites. 'W' and 'C' in brackets indicate chain information, 'W' indicates Watson chain and 'C' indicates Crick chain. The importance of each feature, including sequence context, IPD, and PW, was determined using random forest. Random forest tree analysis showed that the importance of IPD and PW features peaked downstream of the cytosine under investigation, with the main contribution to classification power being that of the IPD and PW downstream of the cytosine under investigation. made it clear that

A random forest consisted of multiple decision trees. During construction of the decision tree, the Gini impurity was used to determine which decision logic of the decision node to use. Important features that have a greater impact on the final classification result are likely to be at nodes closer to the root of the decision tree, while unimportant features that have less impact on the final classification result are more likely to be located further away from the root. It is likely to be in the node where As such, feature importance can be estimated by computing the average distance to the roots of all decision trees in the random forest.

In some embodiments, consensus of methylation calls at CpG sites between Watson and Crick strands can be further used to improve specificity. For example, both strands showing methylation should be called the methylated state, and both strands showing unmethylation should be called the unmethylated state. Since methylation at CpG sites is known to be typically symmetrical, confirmation from each strand can improve specificity.

In various embodiments, global kinetic features from the entire molecule can be used for determination of methylation status. For example, global methylation affects global dynamics during single-molecule real-time sequencing. Modeling the sequencing dynamics of the entire template DNA molecule, including IPD, PW, fragment size, strand information, and sequence context, can improve the accuracy of classification as to whether a molecule is methylated. As an example, the measurement window can be the entire template molecule. IPD, PW, or other kinetic feature statistics (eg, mean, median, mode, percentile, etc.) can be used to determine methylation across the molecule.

B. Limitations of Other Analytical Techniques IPD-based detection of methylation of specific Cs at specific sequence motifs is very low, for example reported to have a sensitivity of only 1.9% (Clark et al., 2013). . We also found that this I tried to reproduce such an analysis. For example, the 3 nt upstream and downstream flanking the CpG under investigation were extracted. The IPDs of that CpG were stratified into different groups (4096 groups for 6 positions) depending on the context of the 6 nt flanking sequences centered on that CpG (ie, 3 nt upstream and downstream, respectively). IPD between methylated and unmethylated CpGs within the same sequence motif was studied using ROC. For example, comparing the IPD of CpGs in the unmethylated 'AATCGGAC' and methylated 'AAT ^m CGGAC' motifs gave an AUC of 0.48. Therefore, using cutoffs in specific sequence groups did not work as well compared to embodiments using various

Figure 27 shows the performance of the motif-based IPD analysis described above for detecting methylation without the use of pulse width signals (Beckmann et al. BMC Bioinformatics. 2014). Vertical bars represent the average AUC (ie, the number of bases surrounding the investigated CpG site) across different k-mer motifs flanking the investigated CpG site. FIG. 27 shows different k-mer motifs (eg, 2-mer, 3-mer, 4-mer, 6-mer, 8-mer, 10-mer, 15-mer, 20-mer surrounding the CpG site of interest). The average AUC for IPD-based discrimination between methylated and unmethylated cytosines over time was found to be less than 60%. These results suggest that considering the IPD of candidate nucleotides in a given motif context, without considering the IPD of neighboring nucleotides (Flusberg et al., 2010), is used here for the determination of CpG methylation. suggested to be inferior to the disclosed method.

We also tested the method present in the study of Flusberg et al. (Flusberg et al., 2010). A total of 5,948,348 DNA segments were analyzed, 2 nts upstream and 6 nts downstream of the cytosine that were subjected to methylation analysis. There were 2,828,848 segments that were methylated and 3,119,500 segments that were unmethylated. As shown in Figure 28, the signals estimated from principal component analysis using IPD and PW were found to have significant overlap between fragments with methylated cytosines (mC) and unmethylated cytosines (C). suggest that the method presented and described by Flusberg et al. lacks any practically meaningful accuracy. These results show that principal component analysis linearly combining PW and IPD values at bases and neighboring bases, as used in the study of Flusberg et al. (Flusberg et al., 2010), showed that suggested that unmethylated cytosines cannot be reliably or meaningfully distinguished.

Figure 29 shows the AUC of the method based on principal component analysis where two principal components were used in Flusberg et al.'s study (Flusberg et al., 2010) including IPD and PW (AUC: 0.55), IPD and PW , as well as the sequence-context-based approach (AUC: 0.94) presented in our disclosure.

C. Other Mathematical/Statistical Models In another embodiment, other mathematical/statistical models including, but not limited to, random forests and logistic regression can be trained by adapting the features developed above. . For the CNN model, the training and test datasets are the M.M. It was constructed from DNA using SssI treatment (methylation) and PCR amplification (unmethylation) (Breiman, 2001). In this random forest analysis, each nucleotide was described using a 4-component binary vector encoding six features: IPD, PW, and base identity . In such a binary vector, A, C, G, and T are [1,0,0,0], [0,1,0,0], [0,0,1,0], and Coded with [0,0,0,1]. For each CpG site analyzed, we incorporated its 10 nt upstream and downstream information on both strands to form a 252-dimensional (252D) vector where each feature represents one dimension. A random forest model as well as a logistic regression model were trained using the training data set described above with 252D vectors. The trained model was used to predict the methylation status of independent test datasets. A random forest consisted of 100 decision trees. Bootstrap samples were used during tree construction. When splitting each decision tree node, the Gini impurity is used to determine the optimal split, and up to 15 features are considered in each split. Also, each leaf of the decision tree had to contain at least 60 samples.

Figures 30A and 30B show the performance of methods using random forest and logistic regression for methylation prediction. FIG. 30A shows the AUC values of training datasets for CNN, random forest, and logistic regression. FIG. 30B shows AUC values for CNN, random forest, and logistic regression test datasets. The method using random forest achieved AUCs of 0.93 and 0.86 on the training and test datasets, respectively.

A logistic regression model was trained using the training data set described with the same 252D vectors. The trained model was used to predict the methylation status of independent test datasets. A logistic regression model (Ng and Y., 2004) with L2 regularization was fitted to the training data set. As shown in FIGS. 30A and 30B, the method using logistic regression achieves AUCs of 0.87 and 0.83 on the training and test datasets, respectively.

These results therefore support the use of specific models other than CNN, such as, but not limited to, random forest and logistic regression, for methylation analysis using the features and analysis protocols developed in this disclosure. suggested that it can be done. These results also show that the CNN implemented according to embodiments of the present disclosure has an AUC of 0.90 on the test dataset (FIG. 30B), random forest (AUC: 0.86) and logistic regression (AUC: 0 .83).

D. Determination of Nucleic Acid 6mA Modifications In addition to methylated CpGs, the methods described herein can also detect other DNA base modifications. For example, methylated adenine can be detected, including the 6mA form.

1. Detection of 6mA Using Kinetic Features and Sequence Context To evaluate the performance and utility of the disclosed embodiments for the determination of base modifications of nucleic acids, we further investigated N6-adenine methylation (6mA ) were analyzed. In one embodiment, about 1 ng of human DNA (eg, extracted from placental tissue) is amplified to produce unmethylated adenine (uA), unmethylated cytosine (C), unmethylated guanine (G), and 100 ng of DNA product was obtained through whole genome amplification with unmethylated thymine (T).

FIG. 31A shows an example of one approach for generating molecules with unmethylated adenines by whole genome amplification. In this figure, "uA" indicates unmethylated adenine and "mA" indicates methylated adenine. Whole-genome amplification is performed using exonuclease-resistant thiophosphate-modified random hexamers as primers, which bind randomly on the genome, allowing a polymerase (e.g., Phi29 DNA polymerase) to amplify the DNA. (e.g. isothermal linear amplification). At stage 3102, the double-stranded DNA is denatured. At stage 3106, the amplification reaction is initiated when a number of random hexamers (eg, 3110) anneal to the denatured template DNA (ie, single-stranded DNA). As shown at 3114, hexamer-mediated DNA synthesis of strand 3118 proceeds in the 5' to 3' direction and upon reaching the next hexamer-mediated DNA synthesis site, the polymerase releases the newly synthesized DNA strand (3122 ) to continue chain elongation. The displaced strand becomes a single-stranded DNA template upon which random hexamers can recombine and initiate new DNA synthesis. Repeated annealing and strand displacement of the hexamers in an isothermal process results in high yields of amplified DNA products. The amplification described herein may fall under the technique of multiple displacement amplification (MDA).

Amplified DNA products can be, for example, but not limited to, , 90 kb, 100 kb, or other desired size range. Fragmentation processes may include enzymatic digestion, nebulization, hydrodynamic shear, sonication, and the like. As a result, original base modifications such as 6mA can be nearly eliminated by whole genome amplification with unmethylated A (uA). FIG. 31A shows possible fragments (3126, 3130, and 3134) of the DNA product, with unmethylated A on both strands. These mA-free whole-genome amplified DNA products were subjected to single-molecule real-time sequencing to generate the uA dataset.

FIG. 31B shows an example of one approach for generating molecules with methylated adenines by whole genome amplification. In this figure, "uA" indicates unmethylated adenine and "mA" indicates methylated adenine. Approximately 1 ng of human DNA was amplified to obtain 10 ng of DNA product through whole genome amplification with 6 mA and unmethylated C, G, and T. Methylated adenines can be produced through a series of chemical reactions (JD Engel et al. J Biol Chem. 1978;253:927-34). As shown in FIG. 31B, whole-genome amplification was performed using exonuclease-resistant thiophosphate-modified random hexamers as primers, which bind randomly on the genome and polymerase ( For example, Phi29 DNA polymerase) is allowed to amplify the DNA (eg, by isothermal linear amplification). Exonuclease-resistant thiophosphate-modified random hexamers are resistant to the 3' to 5' exonuclease activity of proofreading DNA polymerases. Therefore, random hexamers are protected from degradation during amplification.

The amplification reaction was initiated when several random hexamers annealed to the denatured template DNA (ie, single-stranded DNA). When hexamer-mediated DNA synthesis proceeds in the 5' to 3' direction and reaches the next hexamer-mediated DNA synthesis site, the polymerase displaces the newly synthesized DNA strand and continues chain elongation. The displaced strand becomes a single-stranded DNA template and random hexamers recombine to initiate new DNA synthesis. Repeated annealing and strand displacement of the hexamers in an isothermal process results in high yields of amplified DNA products.

The amplified DNA product can be, for example, but not limited to, 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 5 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 60 kb in length. , 70 kb, 80 kb, 90 kb, 100 kb, or other combinations. As shown in Figure 31B, the amplified DNA product will contain different forms of methylation patterns across the adenine sites of each strand. For example, both strands of a double-stranded molecule can be methylated for adenine (molecule I), produced when two strands are derived from DNA synthesis during whole genome amplification.

As another example, one strand of a double-stranded molecule may contain an interlaced methylation pattern across adenine sites (molecule II). An interlaced methylation pattern is defined as comprising a mixture of methylated and unmethylated bases present on a DNA strand. The following example uses an interlaced adenine methylation pattern containing a mixture of methylated and unmethylated adenines present in the DNA strand. This type of double-stranded molecule (molecule II) may be generated because an unmethylated hexamer containing an unmethylated adenine binds to the DNA strand and initiates DNA elongation. Such amplified DNA products containing hexamers with unmethylated adenines will be sequenced. Alternatively, this type of double-stranded molecule (molecule II) is initiated by fragmented DNA from the original template DNA containing unmethylated adenines, since such fragmented DNA serves as a primer This is because it may bind to the DNA strand. Such amplified DNA products containing portions of the original DNA with unmethylated adenines on the strands will be sequenced. Since the unmethylated hexamer primer is only a small portion of the resulting DNA strand, most of the fragments still contain 6mA.

As another example, one strand of a double-stranded DNA molecule may be methylated over an adenine site, while the other strand may be unmethylated (molecule III). This type of double-stranded molecule can be produced when an original DNA strand without methylated adenine is provided as a template DNA molecule for producing a new strand with methylated adenine. .

Both strands are potentially unmethylated (molecule IV). This type of double-stranded molecule may be due to re-annealing of the two original DNA strands that do not have methylated adenines.

Fragmentation processes can include enzymatic digestion, nebulization, hydrodynamic shear, sonication, and the like. Such whole-genome amplified DNA products can be methylated predominantly on the A sites. This DNA with mA was subjected to single-molecule real-time sequencing to generate the mA dataset.

For the uA dataset, 262,608 molecules with a median length of 964 bp were sequenced using single-molecule real-time sequencing. The median sub-read depth was 103-fold. 48% of the subreads could be aligned to the human reference genome using the BWA aligner (Li H et al. Bioinformatics. 2009;25:1754-60). As an example, the Sequel II system (Pacific Biosciences) can be used to perform single-molecule real-time sequencing. Fragmented DNA molecules were subjected to single molecule real-time (SMRT) sequencing template construction using SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences). Conditions for sequencing primer annealing and polymerase binding were calculated using SMRT Link v8.0 software (Pacific Biosciences). Briefly, sequencing primer v2 was annealed to the sequencing template, then polymerase was allowed to bind to the template using the Sequel II Binding and Internal Control Kit 2.0 (Pacific Biosciences). Sequencing was performed on Sequel II SMRT Cell 8M. Sequencing movies were collected for 30 hours on the Sequel II system using the Sequel II Sequencing Kit 2.0 (Pacific Biosciences).

For the mA dataset, 804,469 molecules with a median length of 826 bp were sequenced using single-molecule real-time sequencing. The median sub-read depth was 34-fold. 27% of the subreads could be aligned to the human reference genome using the BWA aligner (Li H et al. Bioinformatics. 2009;25:1754-60).

In one embodiment, kinetic properties including but not limited to IPD and PW were analyzed in a strand-specific manner. Sequence results derived from Watson strands show 644,318 randomly selected A sites containing no methylation from the uA dataset and 718,586 randomly selected methylated A sites from the mA dataset. The A sites were used to construct the training dataset. Using such training data sets, classification models and/or thresholds were established to discriminate between methylated and unmethylated adenines. The test data set consisted of 639,702 A-sites without methylation and 723,320 A-sites with methylation. Such a test dataset was used to validate the performance of the model/threshold estimated from the training dataset.

Sequence results derived from Watson strands were analyzed. FIG. 32A shows pulse interval (IPD) values across the training datasets of the uA and mA datasets. For the training dataset, the IPD values across the sequenced A sites were higher in the mA dataset (median: 1.09, range: 0-9.52) than in the uA dataset (median: 0.20). , range: 0-9.52) (P value < 0.0001, Mann-Whitney U test).

FIG. 32B shows the IPD of the test datasets for the uA and mA datasets. When examining the IPD values across the sequenced A sites of the test dataset, we observed higher IPD values for the mA dataset than for the uA dataset (median 1.10 vs. 0.19, P value <0.0001, Mann-Whitney U test).

FIG. 32C shows the area under the receiver operating characteristic (ROC) curve using the IPD cutoff. The true positive rate is on the y-axis and the false positive rate is on the x-axis. The area under the receiver operating characteristic curve (AUC) in discriminating the sequence A bases of the template DNA molecule with and without methylation using the corresponding IPD values for both the training and test datasets was 0.86.

Sequence results from the Crick strand were analyzed in addition to the results from the Watson strand. FIG. 33A shows the IPD values for the entire training dataset for the uA and mA datasets. For the training dataset, the IPD values across the sequenced A sites were higher in the mA dataset (median: 1.10, range 0-9.52) than in the uA dataset (median: 0.19, range 0-9.52). Range: 0-9.52) was observed (P-value < 0.0001, Mann-Whitney U test).

FIG. 34B shows the IPD values for the test datasets of the uA and mA datasets. Higher IPD values across sequenced A sites were also observed in the mA dataset in the test dataset compared to the uA dataset (median 1.10 vs. 0.19, P value < 0.0001, Mann-Whitney U test).

FIG. 33C shows the area under the ROC curve. The true positive rate is on the y-axis and the false positive rate is on the x-axis. The area under the ROC curve (AUC) values in discriminating the sequenced A bases of the template DNA molecule with and without methylation using the corresponding IPD values are given for the training and test data sets. , were 0.86 and 0.87, respectively.

FIG. 34 shows a diagram of a 6 mA determination of Watson strands using a measurement window, according to an embodiment of the invention. Such measurement windows can include kinetic features such as IPD and PW and nearby sequence context. The 6mA determination can be done similarly to the methylated CpG determination.

FIG. 35 shows a diagram of 6 mA determination of click strands using a measurement window, according to an embodiment of the invention. Such measurement windows can include kinetic features such as IPD and PW and nearby sequence context.

As an example, a measurement window was constructed using 10 bases from each side of the sequenced A bases of the template DNA under investigation. Feature values including IPD, PW, and sequence context were used to train the model using a convolutional neural network (CNN) according to the methods disclosed herein. In other embodiments, statistical models include linear regression, logistic regression, deep recurrent neural networks (e.g., long-term short-term memory, LSTM), Bayesian classifiers, hidden Markov models (HMM), linear discriminant analysis (LDA), It may include, but is not limited to, k-means clustering, density-based spatial clustering for applications with noise (DBSCAN), random forest algorithms, support vector machines (SVM), and the like.

Figures 36A and 36B show the determined probabilities of being methylated for the sequenced A bases of the Watson strand between the uA and mA data sets using the measurement window-based CNN model. FIG. 36A shows the CNN model learned from the training dataset. As an example, the CNN model utilizes two 1D convolutional layers, each with 64 filters with a kernel size of 4 followed by a ReLU layer (regularized linear unit), followed by a dropout rate of 0.5 A dropout layer was used. A maximum pooling layer with a pool size of 2 was used. We then flowed into two 1D convolutional layers (128 filters each with kernel size 2 followed by a ReLU layer) and used a dropout layer with a dropout rate of 0.5. A maximum pooling layer with a pool size of 2 was used. Finally, methylation probabilities were obtained with a fully connected layer containing 10 neurons, followed by a ReLU layer with an output layer containing 1 neuron, followed by a sigmoidal layer. Other settings of layers, filters, kernel sizes can be adapted, eg, as described herein for other methylations (eg, CpG). In this training dataset for Watson strand sequencing results, 644,318 and 718,586 A bases from unmethylated and methylated libraries were used.

Based on the CNN model, for the Watson strand association data, the sequenced A bases of the template DNA molecule from the mA database were compared to those A bases present in uA in the training and test datasets. Both resulted in a much higher probability of methylation (P value <0.0001, Mann-Whitney U test). For the training dataset, the median probability of methylation at the A site for the uA dataset was 0.13 (interquartile range, IQR: 0.09-0.15), whereas for the mA dataset The value was 1.000 (IQR: 0.998-1.000).

FIG. 36A shows the methylation probabilities determined for the test data set. For the test data set, the median probability of methylation at the A site for the uA data set was 0.13 (IQR: 0.10-0.15), whereas the value for the mA data set was 1.000. (IQR: 0.997-1.000). Figures 36A and 36B show that a measurement window-based CNN model can be trained to detect methylation in the test dataset.

FIG. 37 is the ROC curve for detecting 6 mA using the measurement window-based CNN model for the sequenced A bases of the Watson strand. The true positive rate is on the y-axis and the false positive rate is on the x-axis. This figure shows that the AUC values in discriminating sequenced A sites with and without methylation using the CNN model show the training data set composed of the Watson chain sequencing results and the test data. 0.94 and 0.93 respectively for the set. It was suggested that it would be feasible to use the disclosure herein to determine the methylation status of the A site using the Watson chain data. Using a determined methylation probability of 0.5 as a cutoff, a specificity of 99.3% and a sensitivity of 82.6% for the detection of 6mA can be achieved. FIG. 37 shows that the measurement window-based CNN model can be used to detect 6mA with high specificity and sensitivity. The accuracy of the model can be compared with techniques using only the IPD metric.

FIG. 38 shows a performance comparison of IPD metric-based 6mA detection and measurement window-based 6mA detection. Sensitivity is plotted on the y-axis and specificity is plotted on the x-axis. FIG. 38 shows that the performance using the measurement window-based 6mA classification according to the present disclosure (AUC: 0.94) was superior to the conventional method using only the IPD metric (AUC: 0.87). (P value < 0.0001, Delong's test). Measurement window-based CNN models outperformed IPD metric-based detection.

Figures 39A and 39B show the determined probabilities of being methylated for those sequenced A bases in the click strand between the uA and mA datasets using the measurement window-based CNN model. Figure 39A shows the training data set and Figure 39B shows the test data set. Both figures plot the methylation probabilities on the y-axis. Figures 39A and 39B show the sequenced A bases of the template DNA molecule from the mA database compared to those A bases present in the uA database for the click strand association data, based on the CNN model, in the training data. Both the set and the test data set show a much higher probability of methylation (P-value < 0.0001, Mann-Whitney U test).

FIG. 40 shows the performance of 6mA detection using the measurement window-based CNN model for the sequenced A bases of click strands. True positive rate is on the y-axis. False positive rate is on the x-axis. FIG. 40 shows the AUC values in discriminating sequenced A sites with and without methylation using the CNN model, training and test datasets composed of click strand sequencing results. are 0.95 and 0.94, respectively. Performance using the CNN approach disclosed herein (AUC: 0.94) was also shown to be superior to performance using the IPD metric (0.87) alone (P-value < 0.94). 0001). This result suggested that it would be feasible to use the click strand data to determine the methylation status of the A site using the disclosure herein. Using a determined methylation probability of 0.5 as a cutoff, a specificity of 99.3% and a sensitivity of 83.0% for the detection of 6mA can be achieved. FIG. 40 shows that the measurement window-based CNN model can be used to detect 6mA with high specificity and sensitivity.

FIG. 41 shows examples of the methylation status of the entire A base of molecules containing Watson and Crick strands. White dots represent unmethylated adenines. Black dots represent methylated adenines. Dotted horizontal lines represent strands of a double-stranded DNA molecule. Molecule 1 shows that both Watson and Crick strands have been determined to be methylated throughout the A base. Molecule 2 shows that the Crick strand was almost all methylated, whereas the Watson strand was almost all unmethylated. Molecule 3 shows that both the Watson and Crick strands were determined to be almost exclusively methylated across the A bases.

2. Reinforcement Training Using Selective Datasets As shown in Figures 36A, 36B, 39A, and 39B, there was a bimodal distribution of methylation probabilities across the sequenced A bases of the template DNA molecule in the mA data set. rice field. In other words, there were some molecules with uA signals in the mA dataset. This was further evidenced by the presence of fully unmethylated and hemimethylated molecules in the mA dataset (Figure 41). One possible reason is that molecules containing uA in the DNA template still contribute to a significant portion of the mA dataset after whole genome amplification, as molecules containing 6 mA reduce the amplification efficiency of DNA during the whole genome amplification step. It occupies a place. This explanation is that 1 ng of genomic DNA amplified at 6 mA produces only 10 ng of DNA product, whereas 1 ng of genomic DNA amplified with unmethylated A produces 100 ng of DNA product under the same amplification conditions. supported by the fact that it produces Thus, for the mA data set, the original template DNA molecule (Xiao CL et al. Mol 2018;71:306-318), in which adenines are typically unmethylated (eg, 0.051%), represents approximately will account for 10%.

In one embodiment, when trying to train a CNN model to discriminate between mA and uA, selective use of those A bases with relatively high IPD values in the mA data set improves mA detection. reduce the impact of uA data on the training of models for Only A bases with IPD values above a certain cutoff value can be used. A cutoff value may correspond to a percentile. In one embodiment, we will use those A bases of the mA data set that have an IPD value greater than the value at the 10th percentile. In some embodiments, those A with IPD values greater than the value at the 1st, 5th, 15th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, 90th or 95th percentile will be used. . Percentiles may be based on data from all nucleic acid molecules within a reference sample or multiple reference samples.

FIG. 42 shows performance in reinforcement training by selectively using A bases of the mA dataset with IPD values greater than the 10th percentile. FIG. 42 shows the true positive rate on the y-axis and the false positive rate on the x-axis. Using A bases in the mA data set with IPD values greater than the 10th percentile to train the CNN model increased the AUC for discrimination between mA and uA bases to 0.98, with the pre-training IPD values The figure shows that it outperformed the model trained on data without selection by (AUC: 0.94). It was suggested that using IPD values to select mA sites to generate the training data set helps improve discriminatory power.

To further confirm the presence of molecules with uA bases in the mA data set, we found that the 6mA present in the molecule increased the elongation of the polymerase during new strand generation compared to molecules without 6mA. We hypothesized that wells with more subreads would increase the percentage of uA in the mA data set to slow down .

FIG. 43 shows a graph of the number of subreads in each well against the percentage of unmethylated adenine in the mA dataset. The y-axis shows the percentage of uA in the mA data set. The x-axis indicates the number of subreads in each well. After removing A sites that had IPD values below the 10th percentile, the test dataset was reanalyzed using a reinforcement model trained by using mA sites. As the number of subreads per well increases (1 to 10 subreads per sequencing well, 10 to 20 subreads per well, 40 to 50 subreads per well, A gradual increase in uA was observed (ie, increased from 14.6% to 55.05%), including from 60 to 70 and >70 subreads per read. Therefore, wells with a high number of subreads tend to have low mA. Methylation of A can slow the progress of the sequencing reaction. Therefore, sequencing wells with greater depth of subreads are more likely to be unmethylated for A. This behavior can be exploited to detect unmethylated molecules using a cutoff value for the number of subreads associated with a molecule, e.g. can be identified as methylation.

Figure 44 shows the pattern of methyladenines between the Watson and Crick strands of the double-stranded DNA molecules of the test data set. Since the methylation of A is asymmetric, it behaves differently between the two chains. Most molecules are methylated by mA incorporation, leaving some unmethylated A. The y-axis indicates the level of methyladenine in click strands. The x-axis indicates the level of Watson chain methyladenine. Each dot represents a double-stranded molecule. Using a reinforcement model trained with selected mA sites, double-stranded molecules can be classified into different groups according to the methylation level of each strand as follows.
(a) For double-stranded DNA molecules, the levels of methyladenine in the Watson and Crick strands were both greater than 0.8. Such double-stranded molecules were defined as permethylated molecules with respect to the adenine site (Fig. 44, region A). The methyladenine level of a chain was defined as the percentage of A sites determined to be methylated among all A sites of that chain.
(b) For double-stranded DNA molecules, the level of methyladenine in one strand was greater than 0.8, while the other strand was less than 0.2. Such molecules were defined as hemimethylated molecules with respect to the adenine site (Figure 44, regions B1 and B2).
(c) For double-stranded DNA molecules, the levels of methyladenine in the Watson and Crick strands were both less than 0.2. Such double-stranded molecules were defined as fully unmethylated molecules with respect to the adenine site (Fig. 44, region C).
(d) For double-stranded DNA molecules, the levels of methyladenine in Watson and Crick strands did not belong to groups a, b, c. Such double-stranded molecules were defined as molecules with an interlaced methylation pattern with respect to the adenine sites (Figure 44, region D). An interlaced methylation pattern was defined as a mixture of methylated and unmethylated adenines present on the DNA strand.

In some other embodiments, cutoff levels of methyladenine to define unmethylated strands include, but are not limited to, 0.01, 0.05, 0.1, 0.2, 0.3 , 0.4, and less than 0.5. Methyladenine level cutoffs for defining methylated strands include, but are not limited to, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, and 0.99. Exceed.

FIG. 45 is a table showing the percentage of fully unmethylated, hemimethylated, fully methylated, and interlaced methyladenine patterns in the training and test datasets. Molecules in the test data set were fully unmethylated (7.0%), hemimethylated (9.8%), fully methylated (79.4%), and interlaced with methyladenine at the adenine site. Molecules with patterns (3.7%) can be classified. These results are comparable to those shown in the training dataset, with fully unmethylated molecules (7.0%), hemimethylated molecules (10.0%) and fully methylated molecules (79.0%) for the adenine site. 4%), and molecules with an interlaced methyladenine pattern (3.6%).

FIG. 46 shows representative molecular examples of fully unmethylated molecules, hemimethylated molecules, fully methylated molecules, and molecules with interlaced methyladenine patterns with respect to the adenine site. White dots represent unmethylated adenines. Black dots represent methylated adenines. Dotted horizontal lines represent strands of a double-stranded DNA molecule.

In embodiments, increasing the purity of the 6mA base used to train the CNN model can improve performance in discriminating between methylated and unmethylated adenine. To this end, increasing the duration of the DNA amplification reaction to increase the newly generated DNA product can counteract the effect of the unmethylated adenine contributed from the original DNA template. In other embodiments, biotinylated bases can be incorporated during DNA amplification using 6mA. DNA products containing newly generated 6mA can be pulled down and concentrated using streptavidin-coated magnetic beads.

3. Use of 6mA Methylation Profiles 6mA modifications of DNA are present in the genomes of bacteria, archaea, protists and fungi (Didier W et al. Nat Rev Microbiol. 2009;4:183-192). It has also been reported that 6 mA is present in the human genome and accounts for 0.051% of all adenines (Xiao CL et al. Mol Cell. 2018;71:306-318). Given the low content of 6mA in the human genome, in one embodiment, the whole genome amplification step reduces the proportion of 6mA in the dNTP mix (N represents unmodified A, C, G, and T) to By adjusting, a training dataset can be created. For example, ratios of 6mA to dNTPs of 1:10, 1:100, 1:1000, 1:10000, 1:100000, or 1:1000000 can be used. In another embodiment, the adenine DNA methyltransferase M. EcoGII can be used to create a 6mA training data set.

6mA levels were lower in gastric and liver cancer tissues, and this downregulation of 6mA correlated with increased tumorigenesis (Xiao CL et al. Mol Cell. 2018;71:306-318). On the other hand, it has been reported that high levels of 6mA are present in glioblastoma (Xie et al. Cell. 2018; 175:1228-1243). Therefore, the 6 mA approach as disclosed herein will be useful for studying cancer genomics (Xiao CL et al. Mol Cell. 2018;71:306-318; Xie et al. Cell. 2018; 175:1228-1243). Moreover, 6mA was found to be more prevalent and abundant in mammalian mitochondrial DNA and was shown to be associated with hypoxia (Hao Z et al. Mol Cell. 2020; doi:10. 1016/j.molcel.2020.02.018). Therefore, the approach for 6mA detection in the present disclosure will be useful for studying mitochondrial stress responses under different clinical conditions such as pregnancy, cancer, and autoimmune diseases.

IV. Results and Uses A. Detection of Methylation Detection of methylation at CpG sites using the methods described above was performed on various biological samples and genomic regions. As an example, methylation determination using cell-free DNA in the plasma of pregnant women using single-molecule real-time sequencing was validated against methylation determination using bisulfite sequencing. Methylation results can be used for different applications, including determining copy number and diagnosing disorders. The methods described below are not limited to CpG sites and can be applied to any modification described herein.

1. Detection of Methylation of Long DNA Molecules in Placental Tissue Single-molecule real-time sequencing can sequence kilobase-long DNA molecules (Nattestad et al., 2018). Inferring haplotype information of methylation status by deciphering the methylation status of CpG sites using the invention described herein by synergistically exploiting the long-read information of single-molecule real-time sequencing. becomes possible. To demonstrate the feasibility of inferring the methylation status of long reads as well as their haplotype information, 478,739 molecules covered by 28,913,838 subreads were used to analyze placental tissue DNA. Sequencing was determined. There were 7 molecules >5 kb in size. Each was covered, on average, by 3 subreads.

FIG. 47 shows methylation states along a long DNA molecule (ie, haplotype block) of size 6,265 bp. It was sequenced in ZMW with ZMW hall number m54276_180626_162240/40763503 and mapped to the genomic position chr1:113246546-113252811 in the human genome. "-" represents non-CpG nucleotides. "U" represents the unmethylated state of the CpG site. "M" represents the methylation status of the CpG site. Regions 4710 highlighted in yellow indicate CpG island regions commonly known to be unmethylated (Fig. 47). The majority of CpG sites in that CpG island were assumed to be unmethylated (96%). In contrast, 75% of CpG sites outside CpG islands were predicted to be unmethylated. These results suggest that the methylation level outside the CpG island (eg, the shore/shelf of the CpG island) is higher than that of the CpG island. The mixture of methylated and unmethylated states in the haplotype arrangement at regions outside the CpG island indicates variability in methylation patterns. Such observations were generally consistent with current understanding (Zhang et al., 2015; Feinberg and Irizarry, 2010). Thus, this disclosure allows calling out different methylation states along long chains, including methylation and unmethylation states, and that haplotype information of methylation states can be graded. means. Haplotype information refers to the association of the methylation status of CpG sites to contiguous stretches of DNA.

In one embodiment, this approach can be used herein to analyze methylation status along haplotypes to detect and analyze imprinted regions. Imprinted regions are subject to epigenetic regulation resulting in methylation status in a parental manner. For example, one key imprinted region is located on human chromosome 11p15.5 and contains the imprinted genes IGF2, H19, and CDKN1C (P57 ^kip2 ), which are potent regulators of fetal growth (Brioude et al. , Nat Rev Endocrinol. 2018; 14:229-249). Genetic and epigenetic abnormalities in imprinted regions may be associated with disease. Beckwith-Wiedemann syndrome (BWS) is an overgrowth syndrome in which patients are often associated with macroglossia, abdominal wall defect, hemihypertrophy, enlarged abdominal organs, and an increased risk of fetal tumors in early childhood. BWS is believed to result from genetic or epigenetic defects within the 11p15.5 region (Brioude et al, Nat Rev Endocrinol. 2018; 14:229-249). A region called ICR1 (imprint control region 1) located between H19 and IGF2 is variably methylated on the paternal allele. ICR1 induces parental origin-specific expression of IGF2. Thus, genetic and epigenetic abnormalities in ICR1 lead to aberrant expression of IGF2, one of the possible reasons for BWS. Therefore, detection of methylation status along imprinted regions is clinically important.

We downloaded data for 92 imprinted genes from a curated public database of currently reported imprinted genes (http://www.geneimprint.org/). 5 kb upstream and downstream regions of these imprinted genes were used for further analysis. Within these regions, 160 CpG islands are associated with these imprinted genes. We obtained 324,248 circular consensus sequences from placental samples. After removing low-quality circular consensus sequences and short regions overlapping CpG islands (e.g., less than 50% of the length of the relevant CpG islands), 9 overlapping CpG islands corresponding to 8 imprinted genes. We obtained two circular consensus sequences.

Figure 48 shows that nine DNA molecules were sequenced by single-molecule real-time sequencing, overlapping imprinted regions including H19, WT1-AS, WT1, DLK1, MEG3, ATP10A, LRRTM1, and MAGI2. It is a table showing The sixth row contained a DNA stretch overlapping the CpG island containing the imprinted region. "U" represents unmethylated cytosine in the CpG context. "M" represents a methylated cytosine in the CpG context. "*" represents CpG sites that were not covered by the sequencing results. "-" represents nucleotides from non-CpG sites. If the molecule overlaps with a single nucleotide polymorphism (SNP), the genotype is shown in brackets. The seventh column shows the methylation status of the whole molecule. According to embodiments present in the present disclosure, a molecule may be referred to as methylated if it is shown that a majority (eg, greater than 50%) of the CpG sites are methylated. otherwise it is said to be unmethylated.

Of the 9 DNA molecules, 5 (55.6%) were called methylated, not far from the expectation that 50% of the DNA molecules are methylated. As shown in the sixth column of the table in FIG. 48, the majority of CpG sites were shown to be either methylated or unmethylated in concert (ie, as methylated haplotypes). One embodiment, according to embodiments present in the present disclosure, a molecule may be referred to as methylated if it is shown that a majority (e.g., greater than 50%) of the CpG sites are methylated. . Otherwise it is said to be unmethylated. Other cutoffs for determining whether a molecule is methylated can be used, including but not limited to at least 10%, 20%, 30%, 40%, 50% of the CpG sites in the molecule analyzed. %, 60%, 70%, 80%, 90% and 100% are considered methylated.

In another embodiment, a molecule comprising analysis of at least one SNP and analysis of at least one CpG site simultaneously is used to determine whether a region is associated with an imprinted region or a known imprinted gene is aberrant. (eg loss of imprint). For purposes of illustration, Figure 49 had the first molecule from the imprinted region having allele "A" and the second molecule from that imprinting region having allele "G". Assuming that the imprinted region was paternally imprinted, the first molecule from the maternal haplotype was fully unmethylated and the second molecule from the paternal haplotype was fully methylated. In one embodiment, such assumptions provide the ground truth of methylation status and allow testing the performance of base modification detection according to embodiments present in the present disclosure.

FIG. 49 shows an example of determination of methylation patterns in imprinted regions. DNA in biological samples was extracted and ligated with hairpin adapters to form circular DNA molecules. Sequence information and base modifications (eg, methylation status of CpG sites) for these circular DNA molecules were unknown. Those circular DNA molecules were subjected to single-molecule real-time sequencing. After the subreads were mapped to the reference genome, the IPD, PW, and sequence context were determined for each subread base derived from those circular DNA molecules. In addition, those molecules were genotyped. The IPD, PW, and sequence context of the measurement window associated with the CG site will be compared to reference kinetic patterns according to embodiments present in the present disclosure to determine the methylation status of each CpG. If two molecules with different alleles show different methylation patterns, one fully unmethylated and the other fully methylated, the genomic region associated with these two molecules will be an imprinted region. . In one embodiment, for example, as shown in FIG. 49, if such a genomic region happens to be a known imprinted region, the methylation pattern of these two molecules would be expected under normal circumstances. matched the methylation pattern (ie, ground truth). It may indicate the accuracy of the method for classification of methylation status according to embodiments present in the present disclosure. In one embodiment, derivation between measured and expected methylation patterns according to embodiments present in the present disclosure will indicate imprinting abnormalities, e.g., loss of imprinting. .

FIG. 50 shows an example of determination of methylation patterns in imprinted regions. In one embodiment, the imprinting pattern can be further determined through analyzing the methylation pattern of that region across a particular family tree. For example, analysis of methylation patterns and allelic information across paternal, maternal genomes, and offspring can be performed. Such pedigrees may further include the genomes of paternal or maternal grandfathers, paternal or maternal grandmothers or other related genomes. In another embodiment, such analysis can be extended to family trio (mother, father and child) datasets in a particular population, e.g. Acquire genetic and genotype information.

As indicated after sorting, both genotype (alleles in boxes) and methylation status can be determined. For each molecule, the methylation pattern at each site can be provided (eg, all methylated or all unmethylated) to identify from which parent the molecule is inherited. Alternatively, methylation density can be determined and one or more cutoffs determine whether a molecule is hypermethylated (e.g., >80% or other %, from one parent) or hypomethylated. (eg, <20% or other %, from the other parent).

2. Detection of Methylation of cfDNA Molecules As another example, cell-free DNA (cfDNA) methylation is also increasingly recognized as an important molecular signal for non-invasive prenatal testing. For example, cfDNA molecules in regions with tissue-specific methylation can be used to determine the proportional contributions from different tissues such as neutrophils, T-cells, B-cells, liver, placenta, etc. in the plasma of pregnant women. (Sun et al., 2015). The feasibility of using plasma DNA methylation in pregnant women to detect trisomy 21 has also been demonstrated (Lun et al., 2013). cfDNA molecules in maternal plasma were fragmented to a median size of 166 bp. This is much shorter than artificially fragmented E. coli DNA, which is approximately 500 bp in size. It has been reported that cfDNA is not randomly fragmented. For example, terminal motifs in plasma DNA that are associated with tissue origins such as placental origin. Such characteristic properties of cell-free DNA provide a very different sequence context than artificially fragmented E. coli DNA. Therefore, it remains unclear whether such polymerase kinetics allow quantitative estimation of methylation levels of typically cell-free DNA molecules. The disclosure in this patent application is applicable, for example, but not limited to, methylation analysis of cell-free DNA in plasma of pregnant women by using a methylation prediction model trained from the tissue DNA molecules described above. be.

Six plasma DNA samples of pregnant women with male fetuses were sequenced using single-molecule real-time sequencing, corresponding to a median of 111,834 CCSs (range: 61,010-503,582) had a median of 30,738,399 subreads (range: 1,431,215 to 105,835,846). Each plasma DNA was sequenced a median of 262 times (range: 173-320). The dataset was generated from DNA prepared by the Sequel I Sequencing Kit 3.0.

To assess the detection of methylation of cfDNA molecules, we used bisulfite sequencing (Jiang et al., 2014) to analyze the methylation of the above six plasma DNA samples of pregnant women. bottom. A median of 66 million paired-end reads were obtained (58-82 million paired-end reads). The median overall methylation was found to be 69.6% (67.1%-72.0%).

Figure 51 shows a comparison of methylation levels deduced by the new approach and conventional bisulfite sequencing. The y-axis is the methylation level predicted according to the embodiments present in this patent application. The x-axis is the methylation level estimated by bisulfite sequencing. A median of 314,675 CpG sites (range: 144,546 to 1,382,568) were analyzed for plasma DNA results generated from single-molecule real-time sequencing. The median percentage of CpG sites predicted to be methylated was 64.7% (range: 60.8-68.5%), similar to the results deduced from bisulfite sequencing. Looked. As shown in FIG. 51, there is a good correlation (r:0. 96, p-value = 0.0023).

Due to the shallow depth of bisulfite sequencing, it may not be robust to estimate the methylation level of each CpG in the human genome (ie, the percentage of sequenced CpGs that are methylated). Instead, we aggregate read signals covering CpG sites in genomic regions where any two consecutive CpG sites are within 50 nt and where the number of CpG sites is at least 10: Methylation levels of some regions with multiple CpG sites were calculated. The percentage of sequenced cytosines out of the total sequenced cytosines and thymines across CpG sites in a region indicated the methylation level of that region. Regions were divided into different groups according to the methylation level of the region. The methylation probabilities predicted by models learned from the previous training dataset (ie, tissue DNA) increased as methylation levels increased (Fig. 52A). These results further suggested the feasibility and validity of using single-molecule real-time sequencing to predict the methylation status of cfDNA molecules in pregnant women. FIG. 52B showed that methylation levels in a 10 Mb genomic window estimated using single-molecule real-time sequencing according to embodiments present in the present disclosure were well corrected with bisulfite sequencing. (r=0.74, p-value<0.0001).

Figure 53 showed that the genomic representation (GR) of the Y chromosome in maternal plasma of pregnant women measured by single-molecule real-time sequencing correlated well with that measured by BS-seq (r = 0.97, P-value = 0.007). These results suggested that single-molecule real-time sequencing would also allow accurate quantification of DNA molecules from non-hematopoietic tissues, such as placenta, which generally have few contributing DNAs. In other words, the present disclosure has demonstrated the feasibility to simultaneously analyze copy number aberrations and methylation status of native molecules without base conversion and amplification prior to sequencing.

3. CpG Block-Based Methods Some embodiments have multiple CpG sites, including, but not limited to, 2, 3, 4, 5, 10, 20, 30, 40, 50, 100 CpG sites. Methylation analysis can be performed on several genomic regions. The size of such genomic regions can be, for example, but not limited to, 50, 100, 200, 300, and 500 nt. The distance between CpG sites in this region can be, for example, but not limited to, 10, 20, 30, 40, 50, 100, 200, 300 nt. In one embodiment, any two consecutive CpG sites within 50 nt may be superimposed to form a CpG block such that the number of CpG sites within this block is 11 or greater. In such block-based methods, multiple regions can be combined into one window represented as a single matrix, effectively processing the regions together.

As an example, the kinetics of all subreads associated with CpG blocks were used for methylation analysis, as shown in FIG. The predicted IPD profiles of the 10 nt flanking upstream and downstream flanking each CpG within the block were artificially aligned to the CpG site to calculate the average IPD profile (Figure 54). The term "projected" means that the subread kinetic signals are aligned to each corresponding CpG site in question. The average IPD profile of CpG blocks was used to train a model (eg artificial neural network, ANN for short) to identify the methylation status of each block. The ANN analysis included an input layer, two hidden layers and an output layer. Each CpG block was characterized by a feature vector of 21 IPD values input to the ANN. The first hidden layer contained 10 neurons with ReLu as the activation function. The second hidden layer contained 5 neurons with ReLu as the activation function. Finally, the output layer contained one neuron with sigmoid as activation function that outputs the probability of methylation. CpG sites with a methylation probability greater than 0.5 were considered methylated, otherwise unmethylated. The average IPD profile can be used to analyze the methylation status of the whole molecule. If a certain number of sites (e.g., 0, 1, 2, 3, etc.) above a threshold are methylated, or if the molecule has a certain methylation density, the entire molecule is considered methylated. can be

The unmethylated and methylated libraries had 9,678 and 9,020 CpG blocks, each containing at least 10 CpG sites. These CpG blocks covered 176,048 and 162,943 CpG sites in unmethylated and methylated libraries. As shown in Figures 55A and 55B, we were able to achieve over 90% overall accuracy in predicting methylation status on both the training and test datasets. However, such embodiments relying on CpG blocks would greatly reduce the number of CpGs that can be evaluated. By definition, the requirement for a minimum number of CpG sites restricts methylation analysis to specific genomic regions (eg analysis of CpG islands).

B. Determining Origin or Disorder Methylation profiles can be used to detect tissue origin or to determine disorder classification. Methylation profile analysis can be used in combination with other clinical data, including imaging, conventional blood panels, and other medical diagnostic information. Methylation profiles can be determined using any method described herein.

1. Determination of Copy Number Abnormalities In this section, we show that SMRT is accurate for determining copy number, and thus methylation and copy number profiles can be analyzed simultaneously.

Copy number aberrations have been shown to be revealed by sequencing tumor tissue (Chan (2013)). Here we show that cancer-associated copy number aberrations can be identified by sequencing tumor tissue using single-molecule real-time sequencing. For example, for case TBR3033, there are 589,435 and 1,495,225 consensus sequences for tumor DNA and its paired adjacent non-tumor liver tissue DNA, respectively (minimum number of subreads used to construct each consensus sequence). 5 requirements) were acquired. The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0. In one embodiment, the genome was split in silico into 2 Mb windows. The percentage of consensus sequences mapping to each window was calculated and the genome representation (GR) was obtained at 2 Mb resolution. GR can be determined by the number of reads at a position, normalized by total sequence reads across the genome.

FIG. 56A shows the ratio of GRs between tumor DNA and its paired adjacent non-tumor tissue DNA using single-molecule real-time sequencing. The copy number ratio of tumor DNA and its paired adjacent normal tissue DNA is shown on the y-axis and the genomic bin index for each 2 Mb window covering chromosomes 1-22 is shown on the x-axis. In this figure, regions with GR ratios below the 5th percentile of all 2 Mb windows were classified as having copy number loss versus those with GR ratios above the 95th percentile of all 2 Mb windows. Regions were classified as having copy number gains. Chromosome 13 showed a decrease in copy number, while chromosome 20 showed an increase in copy number. Such increases and decreases are correct results.

FIG. 56B shows the ratio of GRs between tumor and its paired adjacent non-tumor tissue using bisulfite sequencing. The copy number ratio of tumor DNA and its paired adjacent normal tissue DNA is shown on the y-axis and the genomic bin index for each 2 Mb window covering chromosomes 1-22 is shown on the x-axis. The copy number changes identified by single-molecule real-time sequencing in Figure 56A were validated with the concordant bisulfite sequencing results in Figure 56B.

For case TBR3032, there are 413,982 and 2,396,054 consensus sequences for tumor DNA and its paired adjacent non-tumor tissue DNA, respectively (minimum requirement of 5 subreads used to construct each consensus sequence). ) were obtained. In one embodiment, the genome was split in silico into 2 Mb windows. The percentage of consensus sequences that mapped to each window, ie, 2Mb genome representation (GR), was calculated.

FIG. 57A shows the ratio of GRs between tumor DNA and its paired adjacent non-tumor tissue DNA using single-molecule real-time sequencing. The copy number ratio of tumor DNA and its paired adjacent normal tissue DNA is shown on the y-axis and the genomic bin index for each 2 Mb window covering chromosomes 1-22 is shown on the x-axis. In this figure, regions with GR ratios below the 5th percentile of all 2 Mb windows were classified as having copy number loss versus those with GR ratios above the 95th percentile of all 2 Mb windows. Regions were classified as having copy number gains. Chromosomes 4, 6, 11, 13, 16, and 17 show reduced copy numbers, and chromosomes 5 and 7 show increased copy numbers. rice field.

FIG. 57B shows the ratio of GRs between tumor and its paired adjacent non-tumor tissue using bisulfite sequencing. The copy number ratio of tumor DNA and its paired adjacent normal tissue DNA is shown on the y-axis and the genomic bin index for each 2 Mb window covering chromosomes 1-22 is shown on the x-axis. The copy number changes identified by single-molecule real-time sequencing in Figure 57A were validated with the concordant bisulfite sequencing results in Figure 57B.

Therefore, methylation profiles and copy number profiles can be analyzed simultaneously. In this example, since the tumor purity of tumor tissue is generally not always 100%, the amplified regions have a relatively increased tumor DNA contribution, and the deleted regions have a relatively increased tumor DNA contribution. relatively decrease. Since tumor genomes are characterized by global hypomethylation, the amplified regions have even lower levels of methylation compared to the deleted regions. As an illustration, for case TBR3033, the methylation level of chromosome 22 (increase in copy number) measured using the present invention was 48.2% and the methylation of chromosome 3 (decrease in copy number) was 48.2%. level (methylation level: 54.0%). For case TBR3032, the chromosome 5p arm methylation level (copy number increase) measured using the present invention was 46.5%, and the chromosome 5q arm methylation level (copy number decrease) (methylation level: 54.9%).

式中、

は、ＤＮＡ混合物中のＣｐＧ部位ｉのメチル化密度を表し、ｐ_ｋは、ＤＮＡ混合物に対する細胞型ｋの比例的な寄与を表し、Ｍ_ｉｋは、細胞型ｋのＣｐＧ部位ｉのメチル化密度を表す。部位の数が臓器の数と同じかそれより多い場合、個々のｐ_ｋ値を決定することができる。有益性を改善するため、ＣｐＧ部位で、すべての参照組織型にわたってメチル化レベルが小さな変動を示すものを除外した。一実施形態では、特定のＣｐＧ部位のセットを使用して、分析を実施した。例えば、様々な組織にわたるメチル化レベルの変動係数（ＣＶ）が３０％を超えていること、および組織間の最大メチル化レベルと最小メチル化レベルとの間の差が２５％を超えていることによって、それらのＣｐＧ部位を特徴付けた。一部の他の実施形態では、５％、１０％、２０％、３０％、４０％、５０％、６０％、８０％、９０％、１００％、１１０％、２００％、３００％などのＣＶも使用することができ、５％、１０％、１５％、２０％、２５％、３０％、４０％、５０％、６０％、７０％、８０％、９０％、１００％などを超える組織間の最大メチル化レベルと最小メチル化レベルとの間の差を使用することができる。 2. Plasma DNA Tissue Mapping in Pregnant Women As shown in Figure 58, the accuracy of the methylation analysis allowed the plasma DNA methylation profiles in pregnant women to be mapped to different reference tissues (e.g., liver, neutrophils, lymphocytes, placenta, T cells, (B cells, heart, brain, etc.). Therefore, DNA contributions in the maternal plasma DNA pool from different cell types can be estimated using the following procedure. CpG methylation levels of DNA mixtures (e.g., plasma DNA) determined according to embodiments present in the present disclosure are recorded in vector (X) and reference methylation levels retrieved across different tissues are quantified (without limitation were recorded in a matrix (M) that can be bisulfite-sequenced). The proportional contribution (p) from different tissues to the DNA mixture can be solved by, but not limited to, quadratic programming. Here we describe the estimation of the proportional contributions of different organs to the DNA mixture using mathematical equations. A mathematical relationship between the methylation densities of different sites in a DNA mixture and the methylation densities of corresponding sites in different tissues can be expressed as follows.

During the ceremony,

is the methylation density of CpG site i in the DNA mixture, _pk is the proportional contribution of cell type k to the DNA mixture, and _Mik is the methylation density of CpG site i in cell type k. show. If the number of sites is equal to or greater than the number of organs, individual _pk values can be determined. To improve the informativeness, CpG sites showing small variations in methylation levels across all reference tissue types were excluded. In one embodiment, analysis was performed using a specific set of CpG sites. For example, the coefficient of variation (CV) of methylation levels across different tissues is greater than 30% and the difference between maximum and minimum methylation levels between tissues is greater than 25%. characterized their CpG sites by In some other embodiments, CV such as 5%, 10%, 20%, 30%, 40%, 50%, 60%, 80%, 90%, 100%, 110%, 200%, 300% can also be used, interstitial greater than 5%, 10%, 15%, 20%, 25%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%, etc. can be used.

Additional criteria can be included in the algorithm to improve accuracy. For example, the aggregate contribution of all cell types can be constrained to be 100%. i.e.
Σ _k p _k = 100%
Furthermore, all organ contributions must be non-negative.
P _k ≥ 0, ∀k

Due to biological variations, the observed global methylation pattern may not be exactly the same as that deduced from tissue methylation. In such situations, mathematical analysis is required to determine the most likely proportional contributions of individual tissues. In this regard, the difference between the methylation pattern observed in DNA and the methylation pattern deduced from tissue is indicated by W.

The most likely value for each p _k can be determined by minimizing W, the difference between the observed and predicted methylation patterns. This equation can be solved using mathematical algorithms such as, but not limited to, quadratic programming, linear/nonlinear regression, expectation maximization (EM) algorithms, maximum likelihood estimation algorithms, maximum posterior probability estimation. , and the least squares method can be used.

As shown in Figure 59, using the method of plasma DNA tissue mapping shown in Figure 58, the contribution of placental DNA to the maternal plasma of pregnant women with male fetuses was estimated by Y-chromosome reads. was observed to correlate well with This result suggested the feasibility of using kinetics to trace the tissue of origin of plasma DNA in pregnant women.

3. Quantification of Methylation Levels of Regions This section describes techniques for determining representative levels of methylation of selected genomic regions. This can be done using relatively low-level sequencing. Methylation levels can be determined on a chain-by-strand, molecule-by-molecule, or region-by-region basis using the number of methylation sites and the total number of methylation sites. Methylation levels of various tissues are also analyzed.

Eleven human tissue DNA samples can be sequenced to a median of 30.7 million subreads per sample (range: 9.1-88.6 million) and aligned to the human reference genome (hg19). Subreads for each sample were generated from Pacific Biosciences Single Molecular Real-Time (SMRT) sequencing wells with a median of 3.8 million (range: 1.1-11.5 million), each well aligned to the human reference genome. At least one subread obtained was included. On average, each molecule in an SMRT well was sequenced an average of 9.9 times (range: 6.5-13.4 times). Human tissue DNA samples included 1 maternal buffy coat sample from pregnant subjects, 1 placental sample, 2 hepatocellular carcinoma (HCC) tumor tissues, and two adjacent HCC tissues paired as previously described. Two non-tumor tissues, four buffy coat samples from healthy control subjects (M1 and M2 from male subjects, F1 and F2 from female subjects), and one HCC cell line (HepG2) were included. rice field. Details of the sequencing data summary are shown in FIG.

FIG. 60 shows different tissue groups in the first column and sample names in the second column. "Total subreads" indicates the total number of sequences generated from SMRT wells, including those from Watson and Crick strands. "Mapped subreads" lists the number of subreads that could be aligned to the human reference genome. "Subread mappability" refers to the percentage of subreads that could be aligned to the human reference genome. "Average sub-read depth per SMRT well" indicates the average number of sub-reads generated from each SMRT well. "Number of SMRT wells" refers to the number of SMRT wells that generated detectable subreads. "Mappable wells" indicates the number of wells containing at least one alignable subread. "% mappable wells" is the percentage of wells containing at least one alignable subread.

a) Methylation Level and Pattern Analysis Techniques In one embodiment, the methylation density of a single nucleic acid strand (e.g., DNA or RNA) can be measured, and the number of methylated bases in the strand is Defined as divided by the total number of methylatable bases. This measurement is also called "single-strand methylation level". This single-strand measurement is particularly feasible in the context of the present disclosure, as single-molecule real-time sequencing platforms can obtain sequencing information from each of the two strands of a double-stranded DNA molecule. This is facilitated by the use of hairpin adapters in preparing the sequencing library so that the Watson and Crick strands of the double-stranded DNA molecule are attached in a circular fashion and sequenced together. Become. In fact, this structure allows the partnered Watson and Crick strands of the same double-stranded DNA molecule to be sequenced in the same reaction, thus allowing the matching of the Watson and Crick strands of any double-stranded DNA molecule. The methylation status of complementary sites can be determined individually and compared directly (eg Figures 20A and 20B).

These strand-based methylation analyzes could not be easily achieved with other techniques. Without using the direct methylation analysis methods disclosed in this application, other means would need to be applied to distinguish methylated from unmethylated bases, for example by bisulfite conversion. Bisulfite conversion requires treatment of DNA with sodium bisulfite so that methylated and unmethylated cytosines can be distinguished as cytosines and thymines, respectively. Under the denaturing conditions of many bisulfite conversion protocols, the two strands of a double-stranded DNA molecule dissociate from each other. In many sequencing applications, for example, using the Illumina platform, the bisulfite-converted DNA is then amplified by the polymerase chain reaction (PCR), with the dissociation of the double-stranded DNA into single strands. .

In Illumina sequencing, methylated adapters can be used prior to bisulfite conversion to prepare sequencing libraries without PCR. Using this strategy, each DNA strand of a double-stranded DNA molecule is also randomly selected for bridge amplification in the flow cell. Due to the random nature of sequencing, it is highly unlikely that each strand from the same DNA molecule will be sequenced in the same reaction. Even if two or more sequences read from the same locus are analyzed in the same run, two reads are from each of the partner Watson and Crick strands of one double-stranded DNA molecule, or two There is no easy way to determine which is from two different double-stranded DNA molecules. Such considerations are important because, in certain embodiments of the invention, the two strands of a double-stranded DNA molecule may exhibit different methylation patterns. When single-strand methylation densities of multiple nucleic acid strands (e.g., DNA or RNA) are measured, the concept and formula for "Methylation level of genomic region of interest" in FIG. ' can also be determined.

FIG. 61 shows various methods of analyzing methylation patterns. A double-stranded DNA molecule (X) with unknown sequence and methylation information is ligated with an adapter, forming, in one example, a hairpin-loop structure. As a result, in this example, the two single strands of the DNA molecule, including both Watson strand X (a) and Crick strand X (b), are physically joined in a circle. The methylation status of sites on both Watson and Crick strands can be obtained using methods described in this disclosure (e.g., kinetic, electronic, electromagnetic, optical signals, or other types of using physical signals). Watson and Crick strands of circularized DNA molecules can be examined in the same reaction. After sequencing, adapter sequences are removed.

From the analysis different methylation levels can be determined. In (I) of FIG. 61, the methylation pattern of only single-stranded molecules, such as either X(a) or X(b), can be analyzed. This analysis can be referred to as single-strand methylation pattern analysis. Analysis can include, but is not limited to, determining the methylation state or pattern of a site. In FIG. 61, single-stranded molecule X(a) shows the methylation pattern 5′-UMMUU-3′, where “U” indicates unmethylated sites, “M” indicates methylated sites, On the other hand, its complementary single-stranded molecule X(b) exhibits the methylation pattern 3'-UMUUU-5'. Therefore, X(b) has a different methylation pattern than X(a). The corresponding single-strand methylation levels of X(a) and X(b) are 40% and 20%, respectively.

In contrast, as shown in (II), methylation patterns can be analyzed at the level of single double-stranded DNA molecules (ie, both Watson and Crick strand methylation patterns are considered). This analysis can be referred to as methylation pattern analysis of single-molecule double-stranded DNA. The methylation level of single-molecule double-stranded DNA of this exemplary molecule X is 30%. In one variant of this analysis, kinetic signals from both Watson and Crick strands are combined to analyze modifications. In particular, since methylation of CpG sites is generally symmetrical, kinetic signals from Watson and Crick strands can be combined for the site before determining the methylation state of the site. In some situations, the performance of determining base modifications using the combined kinetic signals from the Watson and Crick strands of the molecule is superior to the performance of using single-strand kinetic signals independently. For example, as shown in FIG. 20B, using the combined kinetic signals from both strands, including the Watson and Crick strands, compared to using the single strands independently (AUC: 0 .85), giving a larger AUC (0.90) in the test data set.

In FIG. 61 (III), the methylation level of the genomic region of interest is determined, and different DNA molecules with different molecular sizes and different numbers of methylatable sites (e.g., CpG sites) contribute to the genomic region of interest. can. This analysis is sometimes referred to as multi-chain methylation level analysis. The term "multi-stranded" can refer to multiple single-stranded DNA molecules, or multiple double-stranded DNA molecules, or any combination thereof. In this example, there are three double-stranded DNA molecules covering the genomic region of interest: molecule 'X', molecule 'Y', and molecule 'Z', each representing an 'a' strand and a 'b' strand. have. The corresponding methylation level of this region is 9/28, or 32%. The size of the genomic region analyzed is 1 nt, 10 nt, 20 nt, 30 nt, 40 nt, 50 nt, 100 nt, 1 knt (kilonucleotides, i.e. 1000 nucleotides), 2 knt, 3 knt, 4 knt, 5 knt, 10 knt, 20 knt, 30 knt, 40 knt, having a size of 50 knt, 100 knt, 200 knt, 300 knt, 400 knt, 500 knt, 1 Mnt (meganucleotides, i.e., one million nucleotides), 2 Mnt, 3 Mnt, 4 Mnt, 5 Mnt, 10 Mnt, 20 Mnt, 30 Mnt, 40 Mnt, 50 Mnt, 100 Mnt, or 200 Mnt can. A genomic region can be a chromosomal arm or the entire genome.

Methylation patterns can also be determined after determining the methylation status of sites within the molecule. For example, in a scenario with three consecutive CpG sites on a single double-stranded DNA molecule, the methylation patterns of each of the Watson and Crick strands are methylated (M), unmethylated (N), and methylation (M) can be revealed. This pattern, eg, MNM for Watson strand, can be referred to as the "methylation haplotype" of Watson strand in this region. Due to the presence of DNA methylation maintenance activity, the methylation patterns of the Watson and Crick strands of a double-stranded DNA molecule can be complementary to each other. For example, if the CpG site of the Watson strand is methylated, the complementary CpG site of the Crick strand may also be methylated. Similarly, the unmethylated CpG site of the Watson strand can be complementary to the unmethylated CpG site of the Crick strand.

In one embodiment, the methylation level of a single DNA molecule can be measured, which measures the number of methylated bases or nucleotides in the molecule versus the number of methylatable bases or nucleotides in that molecule. Defined as divided by the total number. This measurement is also called the "single-molecule methylation level". This single-molecule measurement may be particularly useful in the context of the present disclosure due to the length of long reads possible on single-molecule real-time sequencing platforms. When measuring single-molecule methylation levels of multiple DNA molecules, a "multi-molecular methylation level" can also be determined based on the concepts and formulas of FIG. For example, "methylation level of multiple molecules" can be the average or median methylation level of a single molecule.

In some embodiments, one or more genetic polymorphisms (e.g., single nucleotide polymorphisms (SNPs)) can be analyzed for a DNA molecule along with the methylation status of sites on the molecule, thus Both the genetic and epigenetic information of the molecule are revealed. Such analysis reveals "graded methylation haplotypes" of the analyzed DNA molecule. Graded methylation haplotype analysis is useful, for example, in studying genomic imprinting in maternal plasma and cell-free nucleic acids (which contain a mixture of cell-free DNA molecules with maternal and fetal genetic and epigenetic properties). be.

b) Comparison of Methylation Results Methylation densities at the genome-wide level for the tissues in the table of FIG. 60 were determined using bisulfite sequencing and single-molecule real-time sequencing, as described in this disclosure. be. FIG. 62A shows methylation density quantified by bisulfite sequencing on the y-axis and tissue type on the x-axis. FIG. 62B shows methylation density on the y-axis and tissue type on the x-axis as quantified by single-molecule real-time sequencing as described in this disclosure.

Figure 62A shows methylation densities across different tissues using bisulfite sequencing (i.e., samples were bisulfite converted and then subjected to Illumina sequencing) (Lister et al. Nature. 2009;462:315- 322), HepG2, HCC tumor tissue, matched HCC tumor-adjacent normal liver tissue (ie, adjacent normal tissue), placental tissue, and buffy coat samples. HepG2 showed the lowest methylation level with a methylation level of 40.4%. The buffy coat sample showed the highest methylation level with a methylation level of 76.5%. The average methylation density of HCC tumor tissue (51.2%) was found to be lower than that of matched adjacent normal tissue (71.0%). This is consistent with the expectation that HCC tumors are hypomethylated at the genome-wide level compared to adjacent normal tissue (Ross et al. Epigenomics. 2010; 2:245-69). . The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

A portion of the same tissue was subjected to single-molecule real-time sequencing and methylation analysis using methods according to the present disclosure. The results are shown in Figure 62B. From methylation analysis using the single-molecule real-time sequencing method of the present disclosure, the HepG2 cell line was the most hypomethylated, followed by the analyzed HCC tumor tissue, followed by the placental tissue. was able to show Adjacent non-neoplastic liver tissue samples were more methylated than other tissues, including HCC and placental tissue, with the buffy coat being the most hypermethylated.

Figures 63A, 63B, and 63C show the correlation of global methylation levels quantified by bisulfite sequencing and single-molecule real-time sequencing according to the methods described herein. FIG. 63A shows methylation levels quantified by bisulfite sequencing on the x-axis and methylation levels quantified by single-molecule real-time sequencing using the methods described herein on the y-axis. . The solid black line is the fitted regression line. The dashed line is where the two measurements are equal.

There was a very high correlation between methylation levels between bisulfite sequencing and single-molecule real-time sequencing according to the invention disclosed herein (r=0.99, P-value<0. 0001). These data demonstrate that methylation analysis using the single-molecule real-time sequencing method disclosed herein is an effective tool for determining methylation levels between tissues, and that It was shown that it enabled the comparison between methylation status and methylation profile. It was noted that the slope of the regression line in FIG. 63A deviated from 1 for the two measurements of methylation level. These results indicate that there is a deviation between the two measurements (in some contexts this deviation may be referred to as a bias), and compared with conventional massively parallel bisulfite sequencing, a single suggested that it may be present in determining methylation levels using molecular real-time sequencing.

との間の関係を使用することができ、以下の式（１）および（２）によって定義された。

式中、「Ｓ」は、本発明による単一分子リアルタイム配列決定によって決定されるメチル化レベルを表し、「バイサルファイトベースのメチル化」は、バイサルファイト配列決定によって決定されるメチル化レベルを表す。 In one embodiment, linear or LOESS (locally weighted smoothing) regression can be used to quantify bias. As an example, if massively parallel bisulfite sequencing (Illumina) is taken as a reference, results determined in single-molecule real-time sequencing according to the present disclosure can be transformed using regression coefficients and between different platforms. You can adjust the readout with . In FIG. 63A, the linear regression equation is Y=aX+b, where “Y” represents the methylation level determined by single-molecule real-time sequencing according to the present disclosure, “X” is the bisulfite sequence Represents the methylation level determined by the assay, where 'a' represents the slope of the regression line (e.g. a=0.62) and 'b' represents the y-axis intercept (e.g. b=17 .72). In this case, the adjusted methylation value determined by single-molecule real-time sequencing is calculated by (Yb)/a. In another embodiment, the deviation (ΔM) of the two measurements and the corresponding average of the two measurements

can be used, defined by equations (1) and (2) below.

where "S" represents methylation levels determined by single-molecule real-time sequencing according to the present invention, and "bisulfite-based methylation" represents methylation levels determined by bisulfite sequencing. .

との間の関係を示す。２つの測定値の平均

は、ｘ軸にプロットされ、２つの測定値間の偏差（ΔＭ）は、ｙ軸にプロットされる。破線は、水平にゼロを横切る線を表し、データポイントは、２つの測定値間に差がないことを示唆している。これらの結果は、平均値に応じて偏差が異なることを示唆した。２つの測定値の平均が高いほど、偏差の大きさが大きくなる。ΔＭ値の中央値は－８．５％（範囲：－１２．６％～＋２．５％）であり、方法間に不一致が存在することを示唆している。 FIG. 63B shows ΔM and

indicates the relationship between Average of two measurements

is plotted on the x-axis and the deviation (ΔM) between the two measurements is plotted on the y-axis. The dashed line represents a line crossing zero horizontally and the data points suggest no difference between the two measurements. These results suggested different deviations depending on the mean. The higher the average of the two measurements, the greater the magnitude of the deviation. The median ΔM value was −8.5% (range: −12.6% to +2.5%), suggesting that there is discrepancy between methods.

をｘ軸に、相対偏差（ＲＤ）をｙ軸に示す。相対偏差は、以下の式によって定義される。

破線は、水平にゼロを横切る線を表し、データポイントは、２つの測定値間に差がないことを示唆している。これらの結果は、相対偏差が平均値に応じて異なることを示唆した。２つの測定値の平均が大きいほど、相対偏差の大きさが大きくなる。ＲＤ値の中央値は、－１２．５％であった（範囲：－１８．１％～＋６．０％）。 Figure 63C is the average of two measurements

is shown on the x-axis and the relative deviation (RD) on the y-axis. Relative deviation is defined by the following equation.

The dashed line represents a line crossing zero horizontally and the data points suggest no difference between the two measurements. These results suggested that the relative deviations differed according to mean values. The greater the average of the two measurements, the greater the magnitude of the relative deviation. The median RD value was −12.5% (range: −18.1% to +6.0%).

Conventional whole-genome bisulfite sequencing (Illumina) has shown, in specific genomic regions, considerable variability in quantification of methylation levels between methods, resulting in significantly biased sequence output and overestimated global methylation. (Olova et al. Genome Biol. 2018; 19:33). The methods disclosed herein can be performed without bisulfite conversion, which dramatically degrades DNA, complicating the process or introducing additional errors in determining methylation levels. It can be performed without possible PCR amplification.

Figures 64A and 64B show methylation patterns at 1 Mb resolution. Figure 64A shows the methylation pattern of the HCC cell line (HepG2). FIG. 64B shows the methylation pattern of buffy coat samples from healthy control subjects. Chromosome ideograms (outermost rings in each figure) are organized clockwise from p-terminus to q-terminus. The second ring from the outside (also called middle ring) shows methylation levels as determined by bisulfite sequencing. The innermost ring shows methylation levels determined by single-molecule real-time sequencing according to the present disclosure. Methylation levels were graded in five grades: 0-20% (light green), 20-40% (green), 40-60% (blue), 60-80% (light red), and 80-100%. (red). As shown in FIGS. 64A and 64B, the methylation profile at 1 Mb resolution is consistent between bisulfite sequencing (middle track) and single-molecule real-time sequencing according to the present disclosure (innermost track). Was. Methylation levels of maternal buffy coat samples were shown to be higher than HCC cell lines (HepG2).

Figures 65A and 65B show scatter plots of methylation levels measured at 1 Mb resolution. FIG. 65A shows methylation levels of HCC cell line (HepG2). FIG. 65B shows methylation levels of buffy coat samples from healthy control subjects. For both Figures 65A and 65B, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single-molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. For HCC cell lines, methylation levels determined by single-molecule real-time sequencing at 1 Mb resolution correlated well with those measured by bisulfite sequencing (r=0.99, P< 0.0001) (Fig. 65A). A correlation was also observed for data from buffy coat samples (r=0.87, P<0.0001) (FIG. 65B).

Figures 66A and 66B show scatter plots of methylation levels measured at 100 kb resolution. FIG. 66A shows methylation levels of HCC cell line (HepG2). FIG. 66B shows methylation levels of buffy coat samples from healthy control subjects. For both FIG. 66A and FIG. 66, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single-molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. A high degree of correlation was also observed between methylation quantification measurements between the two methods at 1 Mb (or 1 Mnt) resolution when the resolution of the analysis was increased by 100 kb (or 100 knt) windows. All these data demonstrate that the single-molecule real-time approach of the present disclosure can detect methylation levels or methylation densities within genomic regions varying with different degrees of resolution, e.g., 1 Mb (or 1 Mnt) or 100 kb (or 100 knt). It has been shown to be an effective tool for quantification. The data also show that the present invention is an effective tool for assessing methylation profiles or patterns between regions or samples.

Figures 67A and 67B show methylation patterns at 1 Mb resolution. Figure 67A shows the methylation pattern of HCC tumor tissue (TBR3033T). Figure 67B shows the methylation pattern of adjacent normal tissue (TBR3033N). Chromosome ideograms (outermost rings in each figure) are organized clockwise from p-terminus to q-terminus. The second ring from the outside (also called middle ring) shows methylation levels as determined by bisulfite sequencing. The innermost ring shows methylation levels determined by single-molecule real-time sequencing according to the present disclosure. Methylation levels were graded in five grades: 0-20% (light green), 20-40% (green), 40-60% (blue), 60-80% (light red), and 80-100%. (red). As shown in Figure 67A, hypomethylation in HCC tumor tissue DNA (TBR3033T) can be detected and can be distinguished from adjacent normal liver tissue DNA (TBR3033N) in Figure 67B. Methylation levels and patterns determined by bisulfite sequencing (middle track) and single-molecule real-time sequencing according to the present disclosure (innermost track) were consistent. The methylation level of adjacent normal tissue DNA was shown to be higher than that of HCC tumor tissue DNA.

Figures 68A and 68B show scatter plots of methylation levels measured at 1 Mb resolution. Figure 68A shows methylation levels of HCC tumor tissue (TBR3033T). FIG. 68B shows methylation levels in adjacent normal tissue. For both Figures 68A and 68B, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single-molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. For HCC tumor tissue DNA, methylation levels measured by single-molecule real-time sequencing at 1 Mb resolution correlated well with those determined by bisulfite sequencing (r=0.96, P value <0.0001) (Fig. 68A). Data from adjacent normal liver tissue samples were also correlated (r=0.83, P-value<0.0001) (FIG. 68B).

Figures 69A and 69B show scatter plots of methylation levels measured at 100 kb resolution. FIG. 69A shows methylation levels of HCC tumor tissue (TBR3033T). Figure 69B shows methylation levels in adjacent normal tissue (TBR3033N). For both Figures 69A and 69B, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. Such a high degree of correlation of methylation quantification data between the two methods at 1 Mb resolution was observed even when measurements of methylation levels were performed at higher resolution, eg, 100 kb windows.

Figures 70A and 70B show the methylation patterns of other tumor and normal tissues at 1 Mb resolution. Figure 70A shows the methylation pattern of HCC tumor tissue (TBR3032T). Figure 70B shows the methylation pattern of adjacent normal tissue (TBR3032N). Chromosome ideograms (outermost rings in each figure) are organized clockwise from p-terminus to q-terminus. The second ring from the outside (also called middle ring) shows methylation levels as determined by bisulfite sequencing. The innermost ring shows methylation levels determined by single-molecule real-time sequencing according to the present disclosure. Methylation levels were graded in five grades: 0-20% (light green), 20-40% (green), 40-60% (blue), 60-80% (light red), and 80-100%. (red). As shown in Figure 70A, we were able to detect hypomethylation in HCC tumor tissue DNA (TBR3032T) and distinguish it from adjacent normal liver tissue DNA (TBR3032N) in Figure 70B. did it. Methylation levels and patterns determined by bisulfite sequencing (middle track) and single-molecule real-time sequencing using the present invention (innermost track) were consistent. The methylation level of adjacent normal tissue DNA was shown to be higher than that of HCC tumor tissue DNA.

Figures 71A and 71B show scatter plots of methylation levels measured at 1 Mb resolution. FIG. 71A shows methylation levels of HCC tumor tissue (TBR3032T). FIG. 71B shows methylation levels in adjacent normal tissue. For both Figures 71A and 71B, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. For HCC tumor tissue DNA, methylation levels measured by single-molecule real-time sequencing at 1 Mb resolution correlated well with those determined by bisulfite sequencing (r=0.98, P <0.0001) (Fig. 71A). Data from adjacent normal liver tissue samples were also correlated (r=0.87, P<0.0001) (FIG. 71B).

Figures 72A and 72B show scatter plots of methylation levels measured at 100 kb resolution. Figure 72A shows methylation levels of HCC tumor tissue (TBR3032T). Figure 72B shows methylation levels in adjacent normal tissue (TBR3032N). For both Figures 72A and 72B, methylation levels quantified by bisulfite sequencing are on the x-axis and methylation levels measured by single-molecule real-time sequencing according to the present disclosure are on the y-axis. The solid line is the fitted regression line. The dashed line is where the two measurement techniques are equal. Such a high degree of correlation of methylation quantification data between the two methods at 1 Mb resolution was observed even when measurements of methylation levels were performed at higher resolution, eg, 100 kb windows.

4. Variable Methylation Regions Between Tumors and Adjacent Normal Tissues Methylomic aberrations are common in regions of cancer genomes. One example of such abnormalities is hypomethylation and hypermethylation of selected genomic regions (Cadieux et al. Cancer Res. 2006;66:8469-76, Graff et al. Cancer Res. 1995;55: 5195-9, Costello et al. Nat Genet. 2000;24:132-8). Another example is the unusual pattern of methylated and unmethylated bases in selected genomic regions. This section shows that techniques that determine methylation can be used to perform quantitative analysis and diagnosis when analyzing tumors.

Figure 73 shows an example of an aberrant pattern of methylation near the tumor suppressor gene CDKN2A. Coordinates are highlighted in blue and underlined indicate CpG islands. Filled dots indicate methylated sites. Unfilled dots indicate unmethylated sites. Numbers in parentheses to the right of each dotted horizontal line indicate fragment size, single-molecule methylation density, and number of CpG sites. For example, (3.3 kb, MD: 17.9%, CG: 39) has a fragment size of 3.3 kb, a fragment methylation level of 17.9%, and a number of CpG sites of 39. means that MD stands for methylation density.

As shown in Figure 73, the CDKN2A (cyclin-dependent kinase inhibitor 2A) gene encodes two proteins that act as tumor suppressors, including INK4A (p16) and ARF (p14). There were two molecules (molecule 7301 and molecule 7302) covering regions that overlapped with the CDKN2A gene in non-tumor tissue adjacent to the tumor tissue. The methylation levels of single double-stranded DNA molecules of molecule 7301 and molecule 7302 were shown to be 17.9% and 7.6%, respectively. In contrast, the methylation level of a single double-stranded DNA molecule of molecule 7303 present in tumor tissue was found to be 93.9%, comparable to the methylation of molecules present in paired adjacent non-tumor tissue. level was much higher. On the other hand, molecules 7301 and 7302 present in non-tumor tissue adjacent to tumor tissue can also be used to calculate multi-chain methylation levels. As a result, the multi-chain methylation level was 9.7%, which was lower than that of tumor tissue (93.9%). Differential methylation levels suggest that single-chain molecule methylation levels and/or multi-chain methylation levels can be used to detect or monitor diseases such as cancer.

Figures 74A and 74B show variable methylated regions detected by single-molecule real-time sequencing, according to embodiments of the invention. FIG. 74A shows hypomethylation in cancer genomes. FIG. 74B shows hypermethylation in cancer genomes. The x-axis indicates the coordinates of the CpG sites. Coordinates are highlighted in blue and underlined indicate CpG islands. Filled dots indicate methylated sites. Unfilled dots indicate unmethylated sites. Numbers in parentheses to the right of each dotted horizontal line indicate fragment size, fragment-level methylation density, and number of CpG sites. For example, (3.1 kb, MD: 88.9%, CG: 180) has a fragment size of 3.1 kb, a fragment methylation density of 88.9%, and a number of CpG sites of 180. means that

FIG. 74A shows regions near the GNAS gene showing more hypomethylated fragments in HCC tumor tissue compared to adjacent normal liver tissue. FIG. 74B shows regions near the ESR1 gene displaying hypermethylated fragments in HCC tissues, whereas DNA fragments from Bear's adjacent non-tumor tissue that align to the corresponding regions instead showed hypomethylation. . As shown in Figure 74B, methylation profiles or methylation haplotypes of individual DNA molecules reveal aberrant methylation status of these genomic regions, GNAS and ESR1, when cancer samples are compared to non-cancer samples. was enough to

These data demonstrate that the single-molecule real-time sequencing methylation analysis disclosed herein determines the methylation status at each CpG site (methylated or unmethylated) on individual DNA fragments. It shows what you can do. Read lengths for single-molecule real-time sequencing are typically much longer (on the order of kilobases) than for Illumina sequencing, which can range from 100-300 nt chain lengths per read (De Maio et al. et al., Micob Genom. 2019;5(9)). Combining the long read length properties of single-molecule real-time sequencing with the methylation analysis methods disclosed herein, methylation haplotypes for multiple CpG sites along any single DNA molecule can be determined. can be easily determined. A methylation profile is the methylation of a CpG site from one coordinate to another in the genome within a continuous stretch of DNA (e.g., on the same chromosome, or within a bacterial plasmid, or within a single DNA stretch within a virus). refers to the state of

Because single-molecule real-time sequencing analyzes each DNA molecule individually without the need for prior amplification, the methylation profile determined for an individual DNA molecule is actually a methylation haplotype, not the same DNA It refers to the methylation status of CpG sites from one end of the molecule to another. If more than one molecule was sequenced from the same genomic region, the % methylation (i.e., methylation level or methylation density) at each CpG site across all sequenced CpG sites within the genomic region is shown in the figure. As shown in 61, the same formula can be used to aggregate data from multiple DNA fragments. The % methylation of each CpG site is reported for all sequenced CpG sites, providing the methylation profile of the sequenced genomic region. Alternatively, data from all reads and all sites within the sequenced genomic region were aggregated, i.e., methylation levels of 1 Mb or 1 kb regions were calculated in the same manner as shown in Figures 64-72. can also provide the 1% methylation value for the region.

5. Viral DNA Methylation Analysis This section demonstrates that the methylation techniques of the present disclosure can be used to accurately determine viral DNA methylation levels.

Figure 75 shows hepatitis B virus DNA methylation patterns between two paired HCC tissue samples and adjacent non-tumor tissue samples using single-molecule real-time sequencing. Each arrow represents a gene annotation of the HBV genome. Arrows with 'P', 'S', 'X', and 'C' indicate gene annotation for the HBV genome, encoding polymerase, surface antigen, X protein, and core protein, respectively. We identified one fragment (molecule I, HBV genome spanning 2,278-3,141 highlighted by dashed rectangle) derived from adjacent non-tumor tissue, 1,183 bp in size. , showed a methylation level of 12%. We also identified three fragments of 3,215 bp, 2,961 bp, and 3,105 bp (molecule II, molecule III, and molecule IV) derived from tumor tissue. Among them, two fragments of HCC tumors (molecule III and molecule IV) overlapped the HBV genomic region spanned by molecule I of non-tumor tissue. In contrast to the hypomethylation level (12%) of the HBV region highlighted by the dashed rectangle (HBV genomic location: 2,278-3,141), methylation levels were significantly higher in those segments of HCC tissue ( was higher (ie, 24% and 30%) for molecules III and IV). These results demonstrate that an approach using single-molecule real-time sequencing is feasible to determine the methylation pattern of the viral genome, and the variable methylation regions of HBV between HCC and non-HCC tissues ( DMR) can be identified. Therefore, determination of the methylation status of the entire viral genome using single-molecule real-time sequencing according to the present disclosure will provide a new tool to study clinical relevance using tissue biopsies.

This DMR region happened to overlap with the P, C, and S genes. This region was also reported to be hypermethylated in HCC tissue compared to cancer-free liver tissue with HBV infection (Jain et al. Sci Rep. 2015;5:10478, Fernandez et al. Genome Res. 2009;19:438-51).

We pooled the bisulfite sequencing results of liver tissue from four patients with cirrhosis but no HCC and obtained 1,156 HBV fragments for methylation analysis. FIG. 76A shows hepatitis B virus DNA methylation levels in liver tissue from patients with cirrhosis but no HCC. In addition, bisulfite sequencing results of HCC tumor tissues from 15 patients were pooled to obtain 736 HBV fragments for methylation analysis. FIG. 76B shows methylation levels of hepatitis B virus DNA in HCC tumor tissue. As shown in FIGS. 76A and 76B, massively parallel bisulfite sequencing also revealed that the DMR region of HBV (HBV genomic location: 1,982-2,435) with higher methylation levels in HCC tissue than in cirrhotic liver tissue. observed. These results suggested that the approach to determine the methylation status of viral genomes is valid.

6. Variant Associated Methylation Analysis Different alleles can be associated with different methylation profiles. For example, an imprinted gene may have one allele with a higher level of methylation than other alleles. In this section, we demonstrate that methylation profiles can be used to identify alleles in specific genomic regions.

One single-molecule real-time sequencing well containing a single DNA template will generate several subreads. Subreads include kinetic characteristics [eg, pulse interval (IPD) and pulse width (PW)] and nucleotide composition. In one embodiment, sub-reads from one single-molecule real-time sequencing well are used to construct a consensus sequence (circular consensus sequence, CCS) that can dramatically reduce sequencing errors (e.g., mismatches, insertions or deletions). ) can be generated. Details of CCS are described herein. In one embodiment, a consensus sequence can be constructed using those subreads aligned to the human reference genome. In another embodiment, consensus sequences can be constructed by mapping subreads to the longest subread within the same single-molecule real-time sequencing well.

Figure 77 shows the principle of stepwise methylation haplotype analysis. Filled lollipops represent CpG sites classified as methylated. Unfilled lollipops represent CpG sites classified as unmethylated.

As shown in one embodiment of Figure 77, the subreads were aligned to the human reference genome. Aligned subreads from one single-molecule real-time sequencing well were combined to form a consensus sequence. A consensus sequence can generally be determined using the most frequent nucleotide present in the subreads across each aligned position. Thus, nucleotide variants, including but not limited to single nucleotide polymorphisms, insertions, and deletions, could be identified from the consensus sequences. Using the averaged IPD and PW within the same molecule tagged with nucleotide variants, the methylation pattern can be determined according to the present disclosure. Thus, variant-associated methylation patterns can be further determined. Methylation states of the same molecule can be considered as methylation haplotypes. Methylation haplotypes are molecular markers that can distinguish whether two or more fragmented short DNA molecules are derived from a single original molecule, or whether two or more different original molecules contribute. Because they do not exist, they may not be readily and directly constructed from two or more short DNA molecules. Synthetic long-read technology (such as the linked-read sequence developed by 10X Genomics) distributes a single long DNA molecule into partitions (such as droplets) that have the same molecular barcode sequence derived from the long DNA molecule. It offers the possibility of tagging short DNA molecules that However, this barcode step involves PCR amplification in which the original methylation state is not preserved.

Furthermore, when using bisulfite to treat long-chain DNA molecules, the first step prior to bisulfite treatment is to: DNA denaturation under destructive conditions that change double-stranded DNA into single-stranded DNA is included. This DNA denaturation step breaks long DNA molecules into shorter fragments and loses the original methylated haplotype information. A second drawback of bisulfite-based methylation analysis is that the bisulfite conversion step denatures double-stranded DNA into single-stranded DNA, the Watson and Crick strands. For one molecule, there is a 50% chance of sequencing the Watson strand and a 50% chance of sequencing the Crick strand. Among the millions of Watson and Crick strands, the probability of sequencing both the Watson and Crick strands of a molecule at the same time is extremely low. Even assuming that both the Watson and Crick strands of a molecule are sequenced, whether such Watson and Crick strands are derived from a single original fragment, or whether two or more different original fragments are It remains impossible to determine with certainty whether or not it will contribute. Liu et al. recently used ten eleven translocation (TET) enzyme-based conversion to develop a bisulfite-free bisulfite-free solution for the detection of methylated and hydroxymethylcytosines under mild conditions that reduce DNA degradation. A sequencing method was introduced (Liu et al. Nat Biotechnol. 2019;37:424-429). However, the enzymatic reaction involves two sequential steps. Low conversion in any step of the enzymatic reaction has a dramatic effect on the overall conversion. Moreover, even with this bisulfite-free sequencing method for detecting methylated cytosines, it is still difficult to distinguish between the Watson and Crick strands of the molecule from the sequencing results.

In contrast, in embodiments of the present invention, the Watson and Crick strands of the molecule are covalently linked via a bell-shaped adapter to form a circular DNA molecule. As a result, both Watson and Crick strands of the molecule can be sequenced in the same reaction well to determine the methylation status of each strand.

One advantage of embodiments of the present invention is the ability to ascertain methylation and genetic (ie, sequence) information on long, contiguous DNA molecules (ekilobases or kilonucleotides in length). Generating such information using short-read sequencing technology is more difficult. For short-read sequencing technologies, to be able to deduce long stretches of methylation and genetic information, it is necessary to combine the sequencing information on multiple short reads using a stepping stone of genetic or epigenetic traits. There is However, this will prove difficult in many scenarios due to the distance between such genetic or epigenetic anchors. For example, there is one SNP per kb on average, but current short-read sequencing technology can typically determine up to 300 nt of sequence per read, 600 nt even in paired-end format.

In one embodiment, variant-associated methylation haplotype analysis can be used to study methylation patterns of imprinted genes. Imprinted regions are subject to epigenetic regulation (eg, CpG methylation) in a parental manner. For example, in the table of Figure 60, one buffy coat DNA sample (M2) was sequenced to obtain approximately 152 million subreads. In this sample, 53% of single-molecule real-time sequencing wells generated at least one subread that could be aligned with the human reference genome. The average sub-read depth of each SMRT well was 7.7 times. In total, approximately 3 million consensus sequences were obtained. Approximately 91% of the reference genome was covered by consensus sequences at least once. For the covered region, the sequencing depth was 7.9 times. The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

Figure 78 shows the size distribution of the sequenced molecules determined from the consensus sequence, with a median size of 6,289 bp (range: 66-198,109 bp). Fragment size (bp) is shown on the x-axis and frequency (%) associated with fragment size is shown on the y-axis.

Figures 79A, 79B, 79C, and 79D show examples of methylation patterns of alleles in imprinted regions. The x-axis indicates the coordinates of the CpG sites. Coordinates are highlighted in blue and underlined indicate CpG islands. Filled dots indicate methylated CpG sites. Unfilled dots indicate unmethylated CpG sites. The alphabet embedded between each horizontal series of filled and unfilled dots (ie, CpG sites) indicates the allele of the SNP site. Numbers in parentheses to the right of each horizontal series of dots indicate fragment size, fragment-level methylation density, and number of CpG sites. For example, (10.0 kb, MD: 79.1%, CG: 139) has a corresponding fragment size of 10.0 kb, a fragment methylation density of 79.1%, and a number of CpG sites of It suggests that there are 139 sites. Dashed rectangles outline the most variable methylation regions within each gene.

Figure 79A shows 11 sequenced fragments with a median size of 11.2 kb (range: 1.3-25 kb) from the SNURF gene. The SNURF gene is maternally imprinted, ie, the copy of the gene that an individual inherits from the mother is methylated and transcriptionally silent. As shown in Figure 79A, in the dashed rectangle, the fragment associated with the C allele was highly methylated, while the fragment associated with the T allele was highly unmethylated. Highly methylated indicates methylation of 70%, 80%, 90%, 95%, or 99% or more of the site. Allele-specific methylation patterns could be observed in other imprinted genes, including PLAGL1 (Fig. 79B), NAP1L5 (Fig. 79C), and ZIM2 (Fig. 79D). FIG. 79B shows that for PLAGL1, the fragment associated with the T allele was highly unmethylated, whereas the fragment associated with the C allele was highly methylated. FIG. 79C shows that for NAP1L5, the fragment associated with the C allele was highly unmethylated, whereas the fragment associated with the T allele was highly methylated. Figure 79D shows that for ZIM2, the fragment associated with the C allele was highly unmethylated, whereas the fragment associated with the T allele was highly methylated.

Figures 80A, 80B, 80C, and 80D show examples of methylation patterns of alleles in non-imprinted regions. The x-axis indicates the coordinates of the CpG sites. Coordinates are highlighted in blue and underlined indicate CpG islands. Filled dots indicate methylated CpG sites. Unfilled dots indicate unmethylated CpG sites. The alphabet embedded between each horizontal series of filled and unfilled dots (i.e., CpG sites) indicates the allele of the single nucleotide polymorphism (SNP site). Numbers in parentheses to the right of the points indicate fragment sizes, fragment-level methylation densities, and number of CpG sites Dashed rectangles are used to calculate methylation densities reported in parentheses , shows randomly selected regions In contrast to the results in Figures 79A-79D, there was no such observable allelic methylation pattern in the non-imprinted genes. Figure 80B shows no difference in the methylation pattern of alleles in the chr7 region, Figure 80B shows no difference in the methylation pattern of the alleles in the chr12 region, Figure 80C shows the methylation of the alleles in the chr1 region. Figure 80D shows no difference in the methylation pattern of the alleles in the different chr1 regions.

Figure 81 shows a table of methylation levels of allele-specific fragments. The first column lists the categories 'imprinted gene' and 'randomly selected region'. The second column lists specific genes. The third column lists the first allele of the SNP of the gene. The fourth column lists the second allele of the SNP of the gene. The fifth column shows the methylation level of the fragment linked to the first allele. The sixth column shows the methylation level of the fragment linked to the second allele. The methylation levels of fragments linked to allele 2 (mean: 88.6%, range 84.6-91.1%) were higher than those of fragments linked to allele 1 of their imprinted genes (mean: 12 .2%, range 7.6-15.7%) (P-value=0.03), indicating the presence of allele-specific methylation. In contrast, there was no significant change in methylation levels between these randomly selected regions (P-value=1), suggesting no allele-specific methylation.

7. Cell-Free DNA Analysis in Pregnancy This example demonstrates that the methods disclosed herein are applicable to the analysis of cell-free nucleic acids in plasma or serum obtained from pregnant women of at least one fetus. do. During pregnancy, cell-free DNA and RNA molecules from placental cells are found in the maternal circulation. Such placenta-derived cell-free nucleic acid molecules are also referred to as cell-free fetal nucleic acids in maternal plasma or circulating cell-free fetal nucleic acids. Cell-free fetal nucleic acid is present in maternal plasma in a background of maternal cell-free nucleic acid. For example, circulating cell-free fetal DNA molecules exist as rare species in the background of cell-free maternal DNA in maternal plasma and serum.

It is known that genetic or epigenetic means or a combination thereof can be used to distinguish cell-free fetal DNA from cell-free maternal DNA in maternal plasma or serum. Genetically, the fetal genome can differ from the maternal genome by paternally inherited fetal-specific SNP alleles, paternally inherited mutations, or de novo mutations. Epigenetically, the placental methylome is generally hypomethylated compared to that of maternal blood cells (Lun et al. Clin Chem. 2013;59:1583-94). Since the placenta is the major contributor of cell-free fetal DNA, while maternal blood cells are the major contributor of cell-free maternal DNA in the maternal circulation (plasma or serum), cell-free fetal DNA molecules are generally expressed in plasma or hypomethylated compared to cell-free maternal DNA in serum. There are specific genomic loci that are hypermethylated in the placenta compared to maternal blood cells. For example, the promoter and exon 1 regions of RASSF1A are more methylated in placenta than in maternal blood cells (Chiu et al. Am J Pathol. 2007;170:941-950). Therefore, circulating cell-free fetal DNA derived from this RASSF1A locus is hypermethylated compared to circulating cell-free maternal DNA derived from the same locus.

In embodiments, cell-free fetal DNA can be distinguished from cell-free maternal DNA molecules based on the differential methylation status between the two pools of circulating nucleic acids. For example, CpG sites along cell-free DNA molecules were found to be mostly unmethylated, suggesting that the molecule may be of fetal origin. If most of the CpG sites along the cell-free DNA molecule are found to be methylated, the molecule is likely maternal. There are several methods known to those skilled in the art to ascertain whether such molecules are in fact fetal or maternal. One approach is to compare the methylation patterns of sequenced molecules with the known methylation profiles of the corresponding loci of placental or maternal blood cells.

FIG. 82 shows an example for using methylation profiles to determine the placental origin of plasma DNA during pregnancy. Coordinates are highlighted in blue and underlined indicate CpG islands. Filled dots indicate methylated sites. Unfilled dots indicate unmethylated sites. Numbers in parentheses near each dotted horizontal line indicate fragment size, single-molecule methylation density, and number of CpG sites.

As shown in Figure 82, maternal plasma cell-free DNA molecules align to the promoter region of RASSF1A, a region known to be differentially methylated in placental tissue, and using the methods of the invention. If the sequencing data generated by the method is hypermethylated, the molecule probably originated from the fetus or placenta. In contrast, molecules exhibiting hypomethylation are likely derived from maternal background DNA (mainly of hematopoietic origin).

Figure 83 shows an approach for fetal-specific methylation analysis. This approach involves the use of sequenced molecules containing fetal-specific SNP alleles or fetal-specific mutations (eg, paternally inherited or de novo in nature). When such fetal-specific genetic features are identified, the methylation status of bases present in the same cell-free DNA molecule reflects the methylation profile of the cell-free fetal DNA or placental methylome. Plasma cell-free DNA sequencing reveals an allele or mutation not present in the maternal genome (e.g., by analysis of maternal genomic DNA), or known to be transmitted by analysis of paternal DNA or in familial Fetal-specific genetic features can be revealed (eg, by analysis of DNA from the proband).

Methylation of fetal-specific DNA molecules can be determined by analyzing those DNA fragments that have alleles that differ from the homozygous alleles of the maternal genome. Methylation of fetal DNA molecules can be expected to be lower than that of maternal DNA molecules.

As an example, buffy coat DNA and the corresponding placental DNA of one pregnant woman were sequenced to obtain genomic coverage of 59-fold and 58-fold haploids, respectively. We identified a total of 822,409 informative SNPs that were homozygous in the mother and heterozygous in the fetus. Through single-molecule real-time sequencing, we found 2,652 fetal-specific fragments and 24,837 shared fragments (ie, fragments with shared alleles, mostly maternally derived) in maternal plasma (M13160). The fetal DNA fraction was 19.3%. Methylation profiles of these fetal-specific and shared fragments were deduced according to the present disclosure. As a result, the methylation level of the fetal-specific fragment was found to be 57.4%, whereas the methylation level of the shared fragment was found to be 69.9%. This finding was consistent with current findings that methylation levels in fetal DNA are lower than maternal DNA in plasma of pregnant women (Lun et al., Clin Chem. 2013;59:1583-94).

Methylation patterns can be used for diagnostic or surveillance purposes. For example, methylation profiles of maternal plasma samples have been used to determine gestational age (https://www.ncbi.nlm.nih.gov/pubmed/27979959). One application is for quality control steps. Another potential application is monitoring the "biological age" and "chronological age" of pregnancy. This application can be used for preterm birth detection or risk assessment. Other embodiments can be used for analysis of fetal cells in maternal blood. In still other embodiments, such fetal cells can be identified by antibody-based approaches or by selective staining using cell markers (e.g., at the cell surface or within the cytoplasm), or by flow cytometry or Concentration can be by micromanipulation or microdissection or physical methods such as differential flow through a chamber, surface or container.

C. Methylation Detection Using Different Reagents This section shows that methylation techniques are not limited to specific reagent systems.

Methylation analysis was performed using different reagent systems to confirm that the technique could be applied. As an example, SMRT-seq was performed to perform single-molecule real-time sequencing using the Sequel II system (Pacific Biosciences). The sheared DNA molecules were subjected to single molecule real-time (SMRT) sequencing template construction using SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences). Conditions for sequencing primer annealing and polymerase binding were calculated using SMRT Link v8.0 software (Pacific Biosciences). Briefly, sequencing primer v2 was annealed to the sequencing template, then polymerase was allowed to bind to the template using the Sequel II Binding and Internal Control Kit 2.0 (Pacific Biosciences). Sequencing was performed on Sequel II SMRT Cell 8M. Sequencing movies were collected for 30 hours on the Sequel II system using the Sequel II Sequencing Kit 2.0 (Pacific Biosciences). In other embodiments, other chemical reagents and reaction buffers will be used for SMRT-seq. In one embodiment, polymerases will have different kinetic characteristics for incorporating nucleotides along the DNA template strand, depending on their methylation state (Huber et al. Nucleic Acids Res. 2016;44:9881-9890). . In this disclosure, results are generated using sequencing primer v1 unless otherwise stated.

To demonstrate the use of the invention in the present disclosure described herein using different reagents, the inventors used, but are not limited to, Sequel I Sequencing Kit 3.0, RS II, Sequel II Sequencing Kit SMRT-seq data generated based on different sequencing kits including Sequel II Sequencing Kit 2.0 and Sequel II Sequencing Kit 2.0 were analyzed. RS II includes 150,000 ZMW per SMRT cell. Sequel uses 1,000,000 ZMW per SMRT cell. Sequel II uses 8 million ZMW per SMRT cell with two sequencing kits (1.0 and 2.0). Two data sets were included in this analysis. The first dataset was prepared based on DNA after whole genome amplification and represents the unmethylated state. A second type of dataset is the M. Prepared based on DNA after SsssI methyltransferase treatment and represents the methylation status. These data were generated using Sequel Sequencing Kit 3.0 for the Sequel sequencer and Sequel II Sequencing Kit 1.0 and Sequel II Sequencing Kit 2.0 for the Sequel II sequencer. rice field. We therefore acquired three data sets with kinetic profiles generated with different reagents (eg polymerases). Each dataset was split into a training dataset and a test dataset for performance evaluation using the CNN model according to the present disclosure.

1. Measurement windows Figures 84A, 84B, and 84C show whole genome amplification data (unmethylated CpG sites) and M. Figure 3 shows the performance of different measurement window sizes across different reagent kits for SMRT-seq on a training dataset containing Ssssl-treated data (methylated CpG sites). The true positive rate is plotted on the y-axis and the false positive rate is plotted on the x-axis. FIG. 84A shows SMRT-seq data generated based on Sequel Sequencing Kit 3.0. FIG. 84B shows SMRT-seq data generated based on Sequel II sequencing Kit 1.0. FIG. 84C shows SMRT-seq data generated based on Sequel II Sequencing Kit 2.0. In the figure, the signal upstream of the analyzed CpG cytosine site is indicated by "-". Signals downstream of the analyzed CpG cytosine site are indicated by "+". For example, "-6nt" represents a 6nt upstream signal for the CpG cytosine site being analyzed. "+6nt" represents the 6nt downstream signal of the CpG cytosine site being analyzed. "±6nt" indicated inclusion of both a 6nt upstream signal and a 6nt downstream signal of the CpG cytosine site being analyzed (ie, a total of 12nt sequences flanking the CpG cytosine site).

Using a measurement window that includes the signal of the CpG cytosine to be analyzed and the signal 6 nt upstream of that cytosine (designated −6 nt) (e.g., IPD, PW, relative position, sequence composition), as shown in FIG. For the training dataset based on Sequel Sequencing Kit 3.0, an AUC value of 0.50 suggested no discriminatory power in distinguishing methylated CpG cytosines from unmethylated cytosines. However, for training datasets based on Sequel II Sequencing Kit 1.0 and 2.0, the corresponding AUC values were 0.62 (Fig. 84B) and 0.75 (Fig. 84C). These data indicate that there are different kinetic profiles inherent to different reagents used in SMRT-seq. These data demonstrate that the methods disclosed herein are readily adapted for use with different reagents. Moreover, the accuracy of detecting base modifications can potentially be improved by further development of reagents, such as using different polymerases and other chemistries.

As another example, as shown in FIG. 84A, using a measurement window containing the signal 10 bp upstream of the CpG cytosine site (labeled −10 nt), for a training dataset based on Sequel Sequencing Kit 3.0, 0 An AUC value of 0.50 suggested that there was no discriminatory power to distinguish methylated CpG cytosines from unmethylated cytosines. However, for training datasets based on Sequel II Sequencing Kit 1.0 and 2.0, the corresponding AUC values were 0.66 (Fig. 84B) and 0.79 (Fig. 84C), indicating that measurements containing 6 nt upstream signals It was shown to be improved compared to windows. These data confirmed that there are different kinetic profiles inherent to the different reagents used for SMRT-seq. These data demonstrate that the methods disclosed herein are readily adapted for use with different reagents.

Measurement windows with downstream signals, as opposed to measurement windows with upstream signals, can lead to significantly improved classification performance. For example, as shown in FIG. 84A, for a training dataset based on Sequel Sequencing Kit 3.0 using a measurement window containing 6 nt downstream signals (+6 nt) of CpG cytosine sites, the AUC value is 0.94, It was much larger than when using the 6nt upstream signal (AUC: 0.5). For training datasets based on Sequel II Sequencing Kit 1.0 and 2.0, the corresponding AUC values are 0.95 (Fig. 84B) and 0.92 (Fig. 84C), respectively, with a measurement window containing 6 nt upstream shows an improvement compared to These data suggest that kinetic features linked to sequence context improve classification power using, but not limited to, CNN models. These data also allow the disclosure herein to use different reagents and different sequencing conditions (e.g., different polymerases, other chemical reagents, their concentrations and sequencing reaction parameters (e.g., , duration)) is applicable to datasets generated by Similar conclusions are drawn from analyzes using measurement windows containing 10 nt downstream signals of CpG cytosine sites (FIGS. 84A, 84B, and 84C).

In another embodiment, a measurement window can be used that includes the signal on the cytosine being analyzed and the signal both upstream and downstream of that cytosine. For example, as shown in FIGS. 84A, 84B, and 84C, using a measurement window containing 6 nt upstream and 6 nt downstream signals (indicated by ±6 nt), AUC values are calculated using Sequel Sequencing Kit 3.0, Sequel II It was found to be 0.94, 0.95, and 0.92 for training datasets based on Sequencing Kit 1.0 and 2.0, respectively. Using a measurement window containing 10 nt upstream and 10 nt downstream signals (denoted as ±10 nt), the AUC values are calculated for training datasets based on Sequel Sequencing Kit 3.0, Sequel II Sequencing Kit 1.0 and 2.0. , were found to be 0.94, 0.95, and 0.94, respectively. These data suggested that the disclosure herein is broadly applicable to data sets generated with different reagents and different sequencing reaction parameters.

Results obtained from test datasets using different measurement windows across different sequence kits when applying the CNN model trained on the training dataset are shown in FIGS. 85A, 85B, and 85C. The true positive rate is plotted on the y-axis and the false positive rate is plotted on the x-axis. The labeling of the legend is comparable to the labeling used in Figures 84A, 84B, and 84C. FIG. 85A shows SMRT-seq data generated based on Sequel Sequencing Kit 3.0. FIG. 85B shows SMRT-seq data generated based on Sequel II sequencing Kit 1.0. FIG. 85C shows SMRT-seq generated based on Sequel II Sequencing Kit 2.0. All conclusions drawn on the training dataset can be validated on these independent test datasets that were not involved in the training process. Furthermore, among the three independent test datasets, analysis of two datasets (2/3) containing Sequel II Sequencing Kits 1.0 and 2.0 showed 10 nt upstream and 10 nt downstream signals (shown as ±10 nt). ) was shown to be superior to other datasets.

2. Comparison with Bisulfite Sequencing Figures 86A, 86B, and 86C show the correlation of global methylation levels quantified by bisulfite sequencing and SMRT-seq (Sequel II Sequencing Kit 2.0). In Figure 86A, the methylation level as a percentage quantified by SMRT-seq is shown on the y-axis. In Figure 86B, the methylation level as a percentage quantified by bisulfite sequencing is shown on the x-axis. The black line is the fitted regression line. The dashed line is a diagonal line with two equal scales. FIG. 86B shows a Bland-Altman plot. The x-axis shows the average methylation levels quantified by SMRT-seq and bisulfite sequencing according to the present disclosure. The y-axis shows the difference in methylation levels between SMRT-seq according to the present disclosure and bisulfite sequencing (ie, Pacific Biosciences methylation-bisulfite-based methylation). Dashed lines correspond to horizontal lines crossing zero where there is no difference between the two scales. Data points outside the dashed line indicate deviations between scales. FIG. 86C shows percentage changes relative to values quantified by bisulfite sequencing. The x-axis shows the average methylation levels quantified by SMRT-seq and bisulfite sequencing according to the present disclosure. The y-axis shows the percentage difference in methylation level between the two scales relative to the mean methylation level. Dashed lines correspond to horizontal lines crossing zero where there is no difference between the two scales. Data points outside the dashed line indicate deviations between scales.

With respect to FIG. 86A, the linear regression equation is Y=aX+b, where “Y” represents the methylation level as determined by SMRT-seq according to the present disclosure and “X” is determined by bisulfite sequencing. 'a' represents the slope of the regression line (eg, a=1.45) and 'b' represents the intercept of the y-axis (eg, b=−20.98). In this case, the methylation value determined by SMRT-seq is calculated by (Yb)/a. This graph shows that methylation levels determined by SMRT-seq can be converted to methylation levels determined by bisulfite sequencing, Sequel II Sequencing Kit 2 as well as Sequel II Sequencing Kit 1.0. .0 and vice versa.

FIG. 86B is a Bland-Altman plot showing the bias in methylation quantification between SMRT-seq and bisulfite sequencing according to the present disclosure, where the x-axis is the Mean quantified methylation levels are shown and the y-axis shows differences in methylation levels quantified by SMRT-seq and bisulfite sequencing according to the present disclosure. The median difference between the two measurements was -6.85% (range: -10.1 to 1.7%). The median percentage change in methylation levels quantified by the present disclosure relative to values by bisulfite sequencing was −9.96% (range: −14,76 to 3.21%). This difference varies depending on the average value. The greater the average of the two measures, the greater the bias.

Figure 86C shows the same data as Figure 86B, but the difference in methylation levels is divided by the average of the two methylation levels. FIG. 86C also shows that the greater the average of the two measurements, the greater the bias.

The error could be in the bisulfite sequencing and not related to the method using SMRT-seq. Conventional whole-genome bisulfite sequencing (Illumina) has shown, in specific genomic regions, considerable variation in quantification of methylation levels between methods, resulting in significantly biased sequence output and overestimated global methylation. (Olova et al. Genome Biol. 2018; 19:33). Embodiments disclosed herein have several exemplary advantages, can be performed without bisulfite conversion, which dramatically degrades DNA, and can be performed without PCR amplification.

3. Tissue Origin Methylation analysis across various cancer types was performed using single-molecule real-time sequencing (SMRT-seq, Pacific Biosciences) according to an embodiment of the present disclosure. Cancer types used for SMRT-seq include colon cancer (n=3), esophageal cancer (n=2), breast cancer (n=2), renal cell carcinoma (n=2), lung cancer (n=2). ), ovarian cancer (n=2), prostate cancer (n=2), gastric cancer (n=2), and pancreatic cancer (n=1). Their matching adjacent non-tumor tissue was also included in the SMRT-seq. The dataset was generated from DNA prepared by the Sequel II Sequencing Kit 2.0.

Figures 87A and 87B show a comparison of global methylation levels between various tumor tissues and paired adjacent non-tumor tissues. Methylation levels are on the y-axis as percentages. In Figure 87A, methylation levels are quantified by SMRT-seq. In Figure 87B, methylation levels are quantified by bisulfite sequencing. Tissue type (ie, tumor tissue or adjacent non-tumor tissue) is on the x-axis. Different symbols represent tissues of different origin.

Figure 87A shows that the global methylation levels of tumor tissues, including breast, colon, esophageal, liver, lung, ovarian, pancreatic, renal cell, and gastric cancers, were compared with the corresponding non-tumor tissues (respectively). , including breast, colon, esophagus, liver, lung, ovary, pancreas, prostate, kidney, and stomach) (P-value=0.006, Wilcoxon signed-rank test of paired samples). The median difference in methylation levels between tumor and paired non-tumor tissues was −2.7% (IQR: −6.4 to −0.8%).

Figure 84B confirms lower methylation levels in tumor tissue. Thus, these results suggest that methylation patterns across various cancer types and tissues can be accurately determined by SMRT-seq according to the present disclosure, providing early detection, prognosis, and diagnosis of cancer on the basis of tissue biopsy. and broad application of the present disclosure for therapy. Differences in the degree of reduction in methylation levels across various tumor types may suggest that methylation patterns are related to cancer type, and may be useful in determining the tissue of origin of the cancer. can.

D. Enhanced Detection and Other Techniques In some embodiments, analysis of base modifications (eg, methylation) can be performed using one or more of the following parameters: sequence context, IPD and PW. IPDs and PWs can be determined from sequencing reactions without alignment to a reference genome. Aspects of single-molecule real-time sequencing approaches can further enhance the accuracy of determining sequence context, IPD, and PW. One aspect is the ability of a circular consensus sequence to measure a particular location in a sequence template multiple times, thereby providing sequence context, IPD, and PW values based on the mean or distribution of values from these multiple reads. can be measured. In certain embodiments, analysis of base modifications without an alignment process can increase computational efficiency, reduce turnaround time, and reduce the cost of analysis. Embodiments can be implemented without an alignment process. In still other embodiments, an alignment process can be used, and may be preferred, e.g., to confirm the clinical or biological significance of detected base modifications. (e.g., the tumor suppressor is hypermethylated), or the alignment process is used to select subsets of the sequencing data corresponding to specific genomic regions of interest for further analysis. be. For those embodiments in which data from a selected genomic region is desired, these embodiments include the region of interest within the genome, e.g., one or more enzymes or It may involve targeting such regions using enzyme-based methodologies. CRISPR-Cas9 systems may be preferred over PCR-based methods because PCR amplification typically does not preserve information about base modifications in DNA. Analysis of methylation levels of such selected regions (biinformatically [e.g., via alignment] or via methods such as CRISPR-Cas9) can be used to determine tissue origin, fetal defects, pregnancy defects, and can provide information about cancer.

1. Methylation Analysis Using Subreads Without Alignment to Reference Genome In embodiments, methylation analysis can be performed using measurement windows that include kinetic features and sequence context from subreads without alignment to a reference genome. . A consensus sequence 8802 (also known as circular consensus sequence (CCS)) was constructed using subreads derived from zero-mode waveguiding (ZMW), as shown in FIG. Mean kinetic values at each location of CCS, including but not limited to PW and IPD values, were calculated. The sequence context surrounding the CpG site was determined from the CCS based on the upstream and downstream sequences of the CpG site. Thus, the measurement window defined in this disclosure is constructed for training, and includes PW, IPD values, and sequence context according to subreads with CCS-relevant kinetic features. This procedure eliminates the need to align subreads to the reference genome.

To test the principle shown in Figure 88, 601,942 unmethylated CpG sites derived from whole genome amplified DNA and 163,527 derived from CpG methyltransferase (e.g., M.SssI) treated DNA. and methylated CpG sites were used to generate a training data set. Using 546,393 unmethylated CpG sites derived from whole genome amplified DNA and 193,641 methylated CpG sites derived from CpG methyltransferase (e.g., M.SssI) treated DNA, test data created a set. The dataset was generated from DNA prepared by the Sequel II Sequencing Kit 2.0.

As shown in FIG. 89, in one embodiment, sub-read and CCS-related kinetic features and sequence context are used to train a convolutional neural network (CNN) model for determining methylation, resulting in and in the training data set, AUC values of 0.94 and 0.95, respectively, that discriminate between methylated and unmethylated CpG sites can be achieved. Other embodiments may use other neural network models, deep learning algorithms, artificial intelligence, and/or machine learning algorithms.

Setting the methylation probability cutoff to 0.2, it is possible to obtain a sensitivity of 82.4% and a specificity of 91.7% in detecting methylated CpG sites. These results demonstrate that subreads with kinetic features can be used to distinguish between methylated and unmethylated CpG sites without prior alignment to the reference genome.

In another embodiment, kinetic features can also be used with sequence context directly from subreads, without CCS information and without prior alignment to the reference genome, to determine the methylation status across CpG sites. Kinetic features including PW and IPD values at positions spanning 20 nt upstream and 20 nt downstream of CpGs present in subreads were used to train a CNN model for determining methylation status. As shown in FIG. 90, the AUC of the ROC curve using sub-read related kinetic features, according to embodiments of the present disclosure, is 0.70 for detecting methylated CpG sites in the training and test datasets, respectively. and 0.69. These data demonstrate that using embodiments of the present disclosure, it is feasible to infer methylation patterns of DNA molecules using kinetic features associated with subreads, but without prior alignment and construction of consensus sequences. suggested that However, the performance of this embodiment to determine methylation was inferior to the embodiments utilizing a combination of alignment information or consensus sequences as described in this disclosure. Enhanced accuracy in generating subreads and kinetic values is expected to improve the ability to determine base modifications using subreads and their associated kinetic features.

2. Methylation Analysis of Deletion Regions Using Targeted Single-Molecular Real-Time Sequencing The methods described herein can also be applied to analyze one or more selected genomic regions. In one embodiment, the region(s) of interest are first obtained by hybridization methods that allow DNA molecules from the region(s) of interest to hybridize to synthetic oligonucleotides having complementary sequences. can be concentrated. Analysis of base modifications using the methods described herein does not transfer the base modification information of the original DNA molecule to the PCR product, so the target DNA molecule cannot be amplified by PCR prior to sequencing. . Several methods have been developed to enrich for these target regions without PCR amplification.

In another embodiment, the target region(s) can be enriched through the use of the CRISPR-Cas9 system (Stevens et al. PLOS One 2019;14(4):e0215441, Watson et al. Lab Invest 2020; 100:135-146). In one embodiment, the ends of the DNA molecules in the DNA sample are first dephosphorylated so that they are not directly ligated to sequencing adapters. The region(s) of interest are then directed by the Cas9 protein with guide RNA (crRNA) to create a double-strand break. The region(s) of interest flanking the double-strand break are then ligated to sequencing adapters specified by the sequencing platform of choice. In another embodiment, DNA can be treated with an exonuclease such that DNA molecules not bound to Cas9 protein are degraded (Stevens et al. PLOS One 2019;14(4):e0215441). Since these methods do not involve PCR amplification, the original DNA molecule containing the base modifications can be sequenced and the base modifications determined. In one embodiment, this method can be used to target multiple regions that share homologous sequences, such as long interspersed repeats (LINEs). In one example, such assays can be used to analyze circulating cell-free DNA in maternal plasma for the detection of fetal aneuploidy (Kinde et al. PLOS One 2012; 7(7). ): e41162).

Implementing targeted single-molecule real-time sequencing using the CRISPR (clustered and regularly arranged short palindromic repeats)/Cas9 (CRISPR-associated protein 9) system, as shown in FIG. can be done. A DNA fragment (eg, molecule 9102) with a 5′ phosphoryl group (ie, 5′-P) and a 3′ hydroxyl group (ie, 3′-OH) has the 5′-P removed and the 3′-OH removed. It was subjected to the end-blocking process by ligation with dideoxynucleotides (ie, ddNTPs). Therefore, the resulting end-modified molecules (eg, molecule 9104) could not be ligated with adapters for subsequent DNA library preparation. However, end-blocked molecules were subjected to target-specific cleavage mediated by the CRISPR/Cas9 system to introduce 5'-P and 3'-OH ends to the molecule of interest. Such newly cleaved DNA molecules with 5′-P and 3′-OH ends (eg, molecule 9106) can now be ligated with hairpin adapters to form circular molecule 9108. . Unligated adapters, linear DNA, and molecules with only one break were subjected to digestion with exonucleases III and VII. As a result, molecules linked by two hairpin adapters were enriched and subjected to single-molecule real-time sequencing. These target molecules were suitable for base modification analysis (ie, targeted single-molecule real-time sequencing) according to embodiments present in the present disclosure.

As shown in Figure 92, the Cas9 proteins of the CRISPR/Cas9 system consist of guide RNAs, including CRISPR RNA (crRNA, involved in DNA targeting) and transactivating crRNA (tracrRNA, involved in forming a complex with Cas9). (ie, gRNA) (Pickar-Oliver et al. Nat Rev Mol Cell biol. 2019;20:490-507). The curved shape represents the Cas9 protein. It is an enzyme that uses the CRISPR sequence as a guide to recognize and cut a specific strand of DNA that is complementary to a portion of the CRISPR sequence. crRNA was annealed to tracrRNA. In one embodiment, the synthetic single RNA sequence contained both crRNA and tracrRNA sequences, referred to as single guide RNA (sgRNA). A segment of crRNA called the spacer sequence directs the Cas9 protein to recognize and cleave a particular strand of double-stranded DNA (dsDNA) through complementary base-pairing to the target region. In one embodiment, there were no mismatches involved in complementarity between the spacer sequence and the target dsDNA. In another embodiment, complementary base pairing between the spacer sequence and the target dsDNA will allow for mismatches. For example, the number of mismatches can be, but is not limited to, 1, 2, 3, 4, 5, 6, 7, 8, and the like. In one embodiment, the CRISPR sequences are programmable according to cleavage efficiency, specificity, sensitivity, and ability to multiplex different CRISPR/Cas complex designs.

As shown in Figure 93, we designed a pair of CRISPR/Cas9 complexes that target two breaks spanning the Alu element of the human genome. "XXX" indicates the three nucleotides flanking the Cas9 nuclease cleavage site. "YYY" indicates the three corresponding nucleotides complementary to "XXX". 5'-NGG represents the protospacer adjacent motif (PAM) sequence. Other CRISPR/Cas systems may have different PAM sequences and different sequences flanking the Cas nuclease cleavage site. In this figure, the size of the Alu region was 223 bp. There were 1,175,329 Alu regions, each containing homologues of such Alu elements in the human genome. A median of 5 CpG sites were located in this Alu element (range: 0-34). As an example, this design contained a 36 nt crRNA containing a 20 nt spacer sequence. Detailed gRNA sequence information is as follows.

First CRISPR/Cas9 complex to introduce the first cut: (all sequences from 5' to 3')
crRNA: GCCUGUAAUCCCAGCACUUUGUUUUAGAGCUAUGCU
tracrRNA: AGCAUAGCAAGUUAAAAAAAGGCUAGUCCGUUAUCAACUUGAAAAAAGUGGGCACCGAGUCGGUGCUUU

A second CRISPR/Cas9 complex to introduce a second cut:
crRNA: AGGGUCUCGCUCUGUCGCCCCGUUUUUAGAGCUAUGCU
tracrRNA: AGCAUAGCAAGUUAAAAAAAGGCUAGUCCGUUAUCAACUUGAAAAAAGUGGGCACCGAGUCGGUGCUUU

The crRNA molecule was annealed to the tracrRNA (eg, 67nt) to form the backbone of the gRNA. A Cas9 nuclease containing engineered gRNA can cleave both strands of an end-blocked molecule with a target cleavage site with a certain level of specificity. There were 116,184 Alu regions of interest in the human genome that were predicted to be cleaved by the designed CRISPR/Cas9 complex. Therefore, these Alu regions can be ligated to hairpin adapters after target cleavage by the Cas9 complex. These molecules ligated to hairpin adapters can be sequenced by single-molecule real-time sequencing. The methylation pattern of these Alu regions can be determined in a targeted manner. In one embodiment, spacer sequences from two Cas9 complexes can base pair with the same strand (eg, Watson strand or Crick strand) of a double-stranded DNA substrate. In one embodiment, the spacer sequences of two Cas9 complex-derived gRNAs are capable of base-pairing with different strands of a double-stranded DNA substrate. For example, one spacer sequence of the Cas9 complex is complementary to the Watson strand of the double-stranded DNA substrate and the other spacer sequence of the Cas9 complex is complementary to the Crick strand of the double-stranded DNA substrate. and vice versa.

In one embodiment, the DNA molecule ligated to the hairpin adapter was in a circular form that is resistant to exonuclease digestion. Thus, the adapter-ligated DNA product can be treated with exonucleases (eg, exonucleases III and VII) to remove linear DNA (eg, off-target DNA molecules). This step using an exonuclease can further enrich target molecules. The size of the target molecule to be sequenced is the span size between the two cleavage sites introduced by one or more Cas9 nucleases (e.g. , 1000 bp, 2000 bp, 3000 bp, 4000 bp, 5000 bp, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 100 kb, 200 kb, 300 kb, 500 kb, and 1 Mb).

As an example, using Cas9 containing gRNAs targeting the Alu region, we used single-molecule real-time sequencing to identify 187,010 molecules from a human hepatocellular carcinoma (HCC) tumor tissue sample. were sequenced. Among them, 113,491 molecules had targeted cleavage (ie, the on-target cleavage rate was approximately 60.7% of the molecules). The dataset was generated from DNA prepared by the Sequel II Sequencing Kit 2.0. In other words, in this example, the specificity of the cleavage site introduced into the molecule of interest by the Cas9 complex was 60.7%. In other embodiments, the specificity of the cleavage site introduced into the molecule of interest by Cas9 or other Cas complexes varies, including but not limited to 1%, 5%, 10%, 20%, 30%, 40% %, 50%, 60%, 70%, 80%, 90%, and 100%. To determine the methylation status at CpG sites of Alu sequences, IPD values, PW values and sequence context derived from CCS and subreads without alignment to the reference genome were used.

As shown in Figure 94, a similar distribution of methylation was observed between methylation levels determined by bisulfite sequencing and single-molecule real-time sequencing according to the present disclosure. Figure 94 shows histograms of percent methylation density for bisulfite sequencing and single molecule real-time sequencing (Pacific Biosciences). The y-axis shows the percentage of molecules in the sample with a particular methylation density indicated on the x-axis. This result suggested that it would be feasible to determine methylation patterns using Cas9-mediated targeted single-molecule real-time sequencing. The results also suggested that kinetic features associated with subreads, including PW and IPD values, could be used to determine methylation without alignment to the reference genome. As shown in Figure 94, we observed significant amounts of Alu regions exhibiting hypomethylation, which was consistent with previous findings that cancer genomes are demethylated at Alu repeat regions (Rodriguez et al. .Nucleic Acids Res.2008;36:770-784).

FIG. 95 shows on the y-axis the distribution of methylation levels determined by single-molecule real-time sequencing according to the present disclosure and on the x-axis methylation density determined by bisulfite sequencing. As shown in Figure 95, the methylation level of the Alu region was divided into five categories according to the results of bisulfite sequencing: 0-20%, 20-40%, 40-60%, 60-80%, and Classified from 80 to 100%. The methylation levels of the same set of Alu regions were further determined by the model using a measurement window containing the kinetic features and sequence context (y-axis) of each category of Alu regions. The distribution of methylation levels determined by our model gradually increased according to the ascending order of methylation levels across binned categories. Again, these results suggest that Cas9-mediated targeted single-molecule real-time sequencing can be used to determine methylation patterns. Methylation can be determined using kinetic features associated with subreads, including PW and IPD values, without alignment to the reference genome.

In yet another embodiment, other types of CRISPR/Cas systems, including but not limited to Cas12a, Cas3, and other orthologues (e.g., Staphylococcus aureus Cas9) or modified Cas proteins (enhanced Acidaminococcus spp Cas12a ) can be used to perform targeted single-molecule real-time sequencing.

In one embodiment, non-activated Cas9 (dCas9), which lacks nuclease activity, can be used to enrich target molecules without cleavage. For example, a target DNA molecule bound to a complex containing biotinylated dCas9 and target sequence-specific gRNA. Since dCas9 lacks a nuclease, such target DNA molecules may not be cleaved by dCas9. Target DNA molecules can be enriched through the use of streptavidin-coated magnetic beads.

In one embodiment, exonucleases can be used to digest the DNA mixture after incubation with Cas proteins. Exonucleases may degrade Cas protein-unbound DNA molecules, whereas exonucleases may not degrade Cas protein-bound DNA molecules, or may degrade them much less efficiently. Therefore, information about target molecules to which Cas proteins have bound can be further enriched in the final sequencing results.

Figure 96 shows a table of tissues and methylation levels of Alu regions within tissues. Many tissues exhibit methylation levels in the 85-92% range, including the 88%-92% range. HCC tumor and placental tissues showed methylation levels below 80%. As seen in Figure 96, HCC tumors were shown to be frequently hypomethylated in the Alu regions targeted by our design. Thus, methylation determination of Alu regions presented in the present disclosure is useful for cancer detection, staging, during tumor progression or during tumor treatment, using DNA extracted from tumor biopsies or other tissues or cells. Can be used for classification and monitoring.

Placental tissue hypomethylation across the Alu region can be used to perform non-invasive prenatal testing using maternal plasma DNA. For example, a high degree of hypomethylation may indicate a high fraction of fetal DNA in pregnant women. In another example, if a woman is pregnant with a chromosomal aneuploid fetus, the number of Alu fragments derived from the affected chromosomes detected by this approach will increase the number of Alu fragments that are pregnant with a euploid fetus. May be quantitatively different (ie either increased or decreased) from females. Thus, if the fetus has trisomy 21, the number of chromosome 21-derived Alu fragments detected by this approach is increased when compared to women carrying euploid fetuses. there is a possibility. On the other hand, if the fetus has a monosomic chromosome, the number of Alu fragments from that chromosome detected by this approach is reduced when compared to women carrying euploid fetuses. there is a possibility. Determination of the presentation of extra hypomethylation of affected chromosomes (13, 18, or 21) in plasma compared with unaffected chromosomes in pregnancies of normal and abnormal fetuses can be used as a molecular marker to distinguish females with

3. Methylation Analysis of Alu Regions Targeted by the Cas9 Complex for Different Types of Cancers Although the targeted Alu repeats were highly methylated in different tissues, we found that different types of cancers were affected by their Alu We hypothesized that we have different demethylation patterns across repeats. In one embodiment, Cas9-based targeted single-molecule real-time sequencing can be used to analyze methylation patterns and determine different cancer types according to the disclosure present herein.

Figure 97 shows cluster analysis of methylation signals associated with Alu repeats in different types of cancer. Cancer subjects from the TCGA database (www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga) were analyzed using microarray technology (Infinium HumanMethylation450 BeadChip, Illumina Inc) had a methylation state in We analyzed the methylation status across 3,024 CpG sites present on the microarray chip and overlapping the Alu regions targeted by the CRISPR/Cas9 complex. There are some CpGs derived from Alu regions of interest in patients. The methylation level of each CpG was quantified by microarray (also called methylation index or beta value). Hierarchical cluster analysis was performed based on the number of methylation levels at those CpG sites across patients. Thus, patients with similar patterns of methylation levels at their CpG sites are grouped together to form clades. The similarity of methylation patterns across different patients is indicated by the clustering dendrogram height values. In this example the height was calculated according to the Euclidean distance. In other embodiments, other distance metrics are used, including but not limited to Minkowski, Chebyshev, Mahalanobis, Manhattan, Cosine, Correlation, Spearman, Hamming, Jaccard distance, and the like. Height, as used herein, represents the value of the distance metric between clusters and reflects the relationships between clusters. For example, if two clusters were observed to overlap with height x, it was suggested that the distance between those clusters was x (eg, the average distance between all inter-cluster patients).

Using the methylation status of CpG sites, cluster analysis results clustered patients into different and distinct groups according to cancer type. Cancer types include bladder urothelial carcinoma (BLCA), invasive breast carcinosarcoma (BRCA), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), HCC, lung adenocarcinoma (LUAD), gastric gland Included are carcinoma (STAD), cutaneous melanoma (SKCM), and uterine carcinosarcoma (UCS). The number after the cancer type in the figure indicates the patient. Thus, clustering suggests that the Alu repeat methylation signals we selected were beneficial for classifying cancer types, including those not shown in FIG. . In one embodiment, primary and secondary tumors can be distinguished based on methylation patterns in tissue biopsies.

4. Subread Depth and Size Cutoffs This section shows that subread depth and/or size cutoffs can be used to improve the accuracy and/or efficiency of methylation detection. Library preparation may be modified to test the depth or size of a particular subread.

Based on the Sequel II Sequencing Kit 2.0, whole genome amplification or M.I. We analyzed the effect of read depth on the quantification of global methylation levels in test datasets generated from samples after SsssI treatment. Genomic sites covered by subreads with at least a particular cutoff, for example, but not limited to, 1-fold, 10-fold, 20-fold, 30-fold, 40-fold, 50-fold, 60-fold, 70-fold, 80-fold, Investigations were made at 90x, 100x, etc.

FIG. 98A shows the effect of read depth on quantification of global methylation levels in test datasets involving whole genome amplification. Figure 98B shows M. Figure 2 shows the effect of read depth on quantification of global methylation levels in test datasets involving SsssI treatment. The y-axis shows the overall methylation level in percentage. The double axis indicates the sub-read depth. The dashed line indicates the expected global methylation level.

As shown in Figure 98A, for the dataset containing whole genome amplification, global methylation ranged from 5.7% to 5.2%, with 1-fold, 10-fold, 20-fold, 40-fold, The first few cutoffs, such as 50-fold, fell off. Methylation levels gradually stabilized at about 5% with a 50-fold cutoff.

On the other hand, in FIG. 98B, M. For datasets generated from samples after SsssI treatment, global methylation ranged from 70% to 83% with the first few such as 1-fold, 10-fold, 20-fold, 40-fold, 50-fold. increased with a cutoff of . Methylation levels gradually stabilized at about 83% with a cutoff of 50-fold or greater.

In one embodiment, the sub-read depth cutoff can be adjusted to make the performance of base modification analysis acceptable across different applications. In other embodiments, using a slightly relaxed sub-read depth cutoff can obtain more ZMWs (ie number of molecules) suitable for downstream analysis. In yet another embodiment, readouts of methylation levels determined by SMRT-seq according to the present disclosure can be calibrated with a second measurement (e.g., but not limited to BS-seq, digital droplet PCR ( on bisulfite-converted samples), methylation-specific PCR, or methylated cytosine-binding antibodies or other proteins). In another embodiment, the second measurement is BS-seq, digital droplet PCR (on bisulfite converted samples), methylation-specific PCR, or methylation-specific PCR of DNA molecules after whole-genome amplification held at 5mC. CpG binding domain (MBD) is obtained by subjecting it to protein-enriched genome sequencing (MBD-seq). As an example, 5mC-retained whole-genome amplification could be mediated by DNA primase TthPrimPol, polymerase phi29, and DNMT1 (DNA methyltransferase 1).

We analyzed methylation levels across different types of cancer and non-tumor tissues for different sub-read depths. Methylation levels determined by SMRT-seq according to the present disclosure were also compared with BS-seq sequencing results. Using Sequel II Sequencing Kit 2.0, we obtained a median of 43 million sub-reads (interquartile range (IQR): 30-52 million), which yielded a median allowed the generation of 4.6 million circular consensus sequences (CCS), aligned with the human reference genome (IQR: 2.8-5.8 million). Of these samples, 22 were also subjected to established Massively Parallel Bisulfite Sequencing (BS-seq) to determine methylation patterns and a second measurement to compare methylation levels. provide value.

FIG. 99 shows a comparison between global methylation levels determined by SMRT-seq (Sequel II Sequencing Kit 2.0) according to the present disclosure and BS-seq using different sub-read depth cutoffs. Methylation levels as percentages determined by SMRT-seq are shown on the y-axis. Methylation levels as percentages determined by bisulfite sequencing are on the x-axis. The symbols indicate different sub-read depths of 1x, 10x and 30x. The three diagonal lines indicate approximate lines for different sub-read depths.

FIG. 99 shows that when analyzing genomic sites that were covered at least once by subreads (i.e., subread depth cutoff of 1-fold or greater), methylation levels of CpG sites determined by SMRT-seq according to the present disclosure were higher than BS- It was shown to correlate well with that determined by seq (r=0.8, P value <0.0001). These results demonstrate that embodiments present in this disclosure include, but are not limited to colon cancer, colorectal tissue, esophageal cancer, esophageal tissue, breast cancer, non-cancerous breast tissue, renal cell carcinoma, kidney tissue, lung cancer, and lung cancer. It has been suggested that it can be used to measure methylation levels in different tissue types including tissue. We also found that when the sub-read depth cutoff was increased by 10-fold and 30-fold, respectively, the correlations between these two measurements were 0.87 (P-value < 0.0001) and 0.95. (P value < 0.0001) was also observed. In some embodiments, increasing the sub-read depth, or selecting genomic regions that cover more sub-reads, will improve the performance of SMRT-seq-based methylation determinations according to the present disclosure.

FIG. 100 is a table showing the effect of sub-read depth on the correlation of methylation levels between the two measurements by SMRT-seq (Sequel II Sequencing Kit 2.0) and BS-seq. The first column shows the sub-read depth cutoff. The second column shows Pearson's r, the correlation coefficient. The third column shows the number of CpG sites associated with the cutoff, with the number range of sites in brackets.

As shown in Figure 100, the correlation of methylation levels between the two measurements by SMRT-seq and BS-seq varied according to different sub-read depth cutoffs. In one embodiment, to determine the optimal cutoff of subread depth for distinguishing methylated from unmethylated cytosines, a subread depth cutoff and a correlation coefficient between the two measurements (e.g., Pearson's correlation coefficient) can be utilized. FIG. 100 shows that at a subread depth cutoff of 30-fold (i.e., 30-fold or greater), methylation levels measured by SMRT-seq according to the present disclosure showed the highest correlation with results generated by BS-seq. (Pearson's r=0.952). Other embodiments include, but are not limited to, 1x, 10x, 30x, 40x, 50x, 60x, 70x, 80x, 900x, 100x, 200x, 300x, 400x, 500x Sub-read depth cutoffs of 1x, 600x, 700x, 800x, etc. can be used.

The number of CpG sites used for methylation analysis decreases with increasing subread depth cutoff, as shown in FIG. Lower correlation between the two measures of methylation levels (Pearson's r=0.875) at a sub-read depth cutoff of 100-fold compared to a 30-fold sub-read depth cutoff (Pearson's r=0.952). ) was observed. The lower correlation at higher subread cutoffs may be due to the lower number of CpG sites that meet the more stringent subread depth cutoffs. In one embodiment, a trade-off can be considered between sub-read depth requirements and the number of molecules that can be used for methylation analysis. For example, more molecules may be desirable if the goal is to scan the entire genome for methylation patterns. When using targeted SMRT-seq to focus on a specific region, a higher sub-read depth may be desirable to obtain the methylation pattern of that region.

FIG. 101 shows the sub-read depth distribution with respect to fragment size for data generated by Sequel II Sequencing Kit 2.0. The y-axis indicates the sub-read depth and the x-axis indicates the strand length of the DNA molecule. The length of the DNA molecule was estimated from the size of the circular consensus sequence (CCS).

Sub-read depth can affect the performance of methylation determinations using SMRT-seq data, and since sub-read depth is a function of the length of the DNA molecule being sequenced, the size of the DNA molecule is It may be important for obtaining optimal sub-read depth for sample methylation pattern analysis. As shown in FIG. 101, the longer the DNA, the shallower the subread depth. For example, for a population of molecules 1 kb in size, the median sub-read depth was 50-fold. For a population of molecules 10 kb in size, the median sub-read depth was 15-fold.

In one embodiment, as shown in diagram 100, the optimal cutoff for sub-read depth may be at least 30 times, yielding the highest correlation coefficient. To further improve the throughput of molecules that meet the 30-fold optimal subread depth cutoff, the relationship between subread depth and DNA template molecule strand length can be exploited. For example, in FIG. 101, 30× is the median sub-read depth for molecules with chain lengths of approximately 4 kb. Therefore, prior to preparing the SMRT-seq library, 4 kb DNA molecules can be fractionated to limit sequencing to 4 kb DNA molecules. In other embodiments, other size cutoffs can be used for fractionation of DNA molecules, including but not limited to 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 2 kb, 3 kb, 4 kb, 5 kb, 6 kb, 7 kb, 9 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 60 kb, 70 kb, 80 kb, 90 kb, 100 kb, 500 kb, 1 Mb, or combinations with different size cutoffs.

5. Restriction Enzyme-Based Targeted Single Molecule Real-Time Sequencing This section describes the use of restriction enzymes to improve the practicality and/or throughput and/or cost-effectiveness of detection of modifications. Restriction enzyme generated DNA fragments can be used to identify the origin of a sample.

a) Using Restriction Enzymes to Digest DNA Molecules In embodiments, prior to single molecule real-time sequencing (e.g., using a Pacific Biosciences system), DNA molecules are digested using one or more restriction enzymes. can be digested. Because the distribution of restriction enzyme recognition sites is heterogeneous in the human genome, DNA digested by restriction enzymes can produce a skewed size distribution. Genomic regions with more restriction enzyme recognition sites can be digested into smaller fragments, while genomic regions with fewer restriction enzyme recognition sites can be digested into longer fragments. In embodiments, the size range allows selective retrieval of DNA molecules derived from one or more regions with similar cleavage patterns of one or more restriction enzymes. The size range required for size selection can be determined by in silico cleavage analysis of one or more restriction enzymes. A computer program can be used to determine the number of recognition sites for a restriction enzyme of interest in a reference genome (eg, a human reference genome). Such reference genomes were sheared in silico into fragments according to their recognition sites, which provide size information for genomic regions of interest.

Figure 126 shows a method for MspI-based targeted single-molecule real-time sequencing using DNA end repair and A tailing. In embodiments, an MspI that recognizes the 5'C^CGG3' site can be used to digest biological DNA samples, such as, but not limited to, human DNA samples, as shown in Figure 126 . Digested DNA fragments with 5'CG overhangs were subjected to size selection to enrich for DNA molecules derived from CpG islands. Genomic regions enriched in G and C residues (also called GC content) may generate shorter fragments. Thus, one can determine a range of fragment sizes for selection based on the GC content of the region of interest. A variety of DNA fragment size selection tools are available to those skilled in the art, including but not limited to gel electrophoresis, size exclusion electrophoresis, capillary electrophoresis, chromatography, mass spectrometry, filtration approaches, precipitation-based approaches, microfluidics. , and nanofluidics. The size-fractionated DNA molecules were subjected to DNA end-repair and A-tailing, and the desired DNA product was ligated with hairpin adapters with 5'T overhangs to form a circular DNA template.

For example, without limitation, exonucleases (exonucleases III and VII) are used to remove unligated adapters, linear DNA, and incomplete circular DNA, followed by DNA molecules ligated to hairpin adapters. , can be used for single-molecule real-time sequencing to determine the IPD, PW, and sequence context in determining the methylation profiles disclosed herein. DNA obtained from different tissues or tissues or biological samples with different diseases and/or physiological conditions by analyzing genomic regions enriched in CpGs determined by the sequencing data analysis methods of the present disclosure. can be distinguished and classified by their methylation profiles.

In embodiments, for the step involving size selection in FIG. 126, the desired size range can be determined by an in silico cleavage analysis of MspI. A total of 2,286,541 MspI cleavage sites were determined in the human reference. The human reference genome was sheared in silico into fragments according to their MspI cleavage sites. A total of 2,286,565 fragments were obtained. Individual fragment sizes were determined by the total number of nucleotides in that fragment.

Figures 127A and 127B show the size distribution of MspI-digested fragments. The y-axis of these figures is the frequency (percentage) of fragments of a particular size. Figure 127A has a logarithmic scale for the x-axis ranging from 50 to 500,000 bp. Figure 127B has a linear scale for the x-axis ranging from 50 to 1,000 bp.

As shown in Figures 127A and 127B, MspI-digested DNA molecules have a skewed size distribution. The median size of MspI-digested fragments was 404 bp (IQR: 98-1,411 bp). Approximately 53% of those MspI-digested fragments were less than 1 kb. The size profile had a series of spike peaks that could be attributed to repetitive elements. Certain repeat elements are likely to share a similar pattern of MspI cleavage sites, leading to a set of molecules derived from MspI digestion with similar fragment sizes. For example, the highest frequency (ie, 49,079 total) spike peaks corresponded to a size of 64 bp. Among them, 45,894 (94%) overlapped with Alu repeats. A DNA molecule of 64 bp in size can be selected to enrich for DNA molecules derived from Alu repeats. The data suggest that size selection can be used to enrich desired DNA molecules for downstream methylation analysis according to the present disclosure.

Figure 128 shows a table of the number of DNA molecules in certain selected size ranges. The first column indicates the size range in base pairs. The second column shows the percentage of molecules within the size range for all fragments. The third column shows the number of molecules within the size range that overlap with the CpG islands. The fourth column shows the percentage of molecules within the size range that overlap with the CpG islands. The fifth column indicates the number of CpG sites sequenced. The sixth column indicates the number of CpG sites within the CpG island. The seventh column shows the percentage of CpG sites that were size-selected and that are within CpG islands. As shown in Figure 128, the amount of DNA molecules generated from the human genome subjected to MspI digestion varied according to the different size ranges of interest. The number of DNA molecules that overlap CpG islands varies with size range.

Since CCGG motifs occur preferentially in CpG islands, molecules of size below a certain cutoff can be selected to allow enrichment of DNA molecules derived from CpG islands. For example, for the 50-200 bp size range, the number of molecules is 526,543, accounting for 23.03% of the total DNA fragments derived from the human genome that were subjected to MspI digestion. Of the 526,543 DNA molecules, 104,079 (19.76%) overlapped with CpG islands. In the 600-800 bp size range, the number of molecules was 133,927, accounting for 5.86% of the total DNA fragments derived from the human genome subjected to MspI digestion. Of the 133,927 molecules, 3,673 (2.74%) molecules overlapped with CpG islands. As an example, a size of 50-200 bp can be chosen to enrich for DNA fragments derived from CpG islands.

To calculate the enrichment of CpG sites overlapping CpG islands via MspI-based targeted single-molecule real-time sequencing, we performed simulations of DNA sheared by sonication, with a standard deviation of 20 bp based on a normal distribution. , simulated 526,543 fragments generated from ZMW with an average size of 200 bp. Only 0.88% of the DNA molecules overlapped with CpG islands. A total of 71,495 CpG sites overlapped with CpG islands. As shown in Figure 128, selecting MspI digested fragments ranging from 50-200 bp results in 19.8% of the fragments overlapping CpG islands. These data therefore suggest that DNA prepared by MspI digestion may be 22.5-fold enriched for DNA fragments derived from CpG islands compared to DNA prepared by sonication. It suggests. In addition, CpG sites enriched in CpG islands were analyzed through MspI digestion. Selection of MspI digested fragments in the 50-200 bp range could result in 885,041 CpG sites overlapping CpG islands, with 37.5 of the total CpG sites from sequenced fragments within that size range. 5%. CpG sites overlapping CpG islands were enriched 12.3-fold (ie, 885,041/71,495) compared to DNA prepared by sonication. Based on the information presented in Figure 128, a suitable size range can be selected to contain the desired number of CpG sites and the desired enrichment factor of CpG sites within a CpG island.

FIG. 129 is a graph of percent coverage of CpG sites within CpG islands versus DNA fragment size after restriction enzyme digestion. The y-axis shows the percentage of CpG sites within the CpG island covered by fragments with a given size. The x-axis indicates the upper limit of the DNA fragment size range after restriction enzyme digestion. Figure 129 shows the percentage of CpG sites within a CpG island that are covered by widening the size selection range. In Figure 129, the size range is from 50 bp to the size indicated on the x-axis. In other embodiments, the lower end of the size range can be customized, including but not limited to 60bp, 70bp, 80bp, 90bp, 100bp, 200bp, 300bp, 400bp, and 500bp. It can be seen that increasing the size range by increasing the upper limit gradually increases the percent coverage of CpG sites within the CpG islands and levels off at 65%. Some CpG sites are not covered because they are within DNA fragments less than 50 bp or within fragments within very long molecules (eg >100,000 bp).

In some embodiments, DNA samples can be analyzed using two or more different restriction enzymes (with different restriction sites), thus increasing the coverage of CpG sites within the CpG islands. . Digestion of DNA samples with different enzymes can be performed in separate reactions such that only one restriction enzyme is present in each reaction. For example, AccII, which recognizes CG^CG sites, can be used to preferentially cleave CpG islands. In other embodiments, other restriction enzymes containing CG dinucleotides as part of the recognition site can be used. There were 678,669 AccII cleavage sites within the human genome. An in silico digestion of the human reference genome was performed using AccII restriction and a total of 678,693 fragments were obtained. We then size-selected these fragments in silico and calculated the percent coverage of CpG sites within the CpG islands according to the method described above for MspI digestion. It can be seen that the percent coverage of CpG sites gradually increases with increasing size selection. The percent coverage levels off at about 50%. CpG site coverage is further increased by combining data from two enzymatic digestion experiments (ie, MspI and AccII digestion). 80% of the CpG sites within the CpG islands are covered by selecting DNA fragments between 50bp and 400bp in size. This percentage is higher than the respective figures for digestion experiments with either of the two enzymes alone. Coverage can be further enhanced through analysis of DNA samples using other restriction enzymes. If the DNA sample is divided into two aliquots, one aliquot is digested with MspI and the other with AccII. The two digested DNA samples are mixed at equimolar concentrations and sequenced using single-molecule real-time sequencing with 5 million ZMW. Based on in silico analysis, 83% of the CpG sites within a CpG island (ie, 1,734,345 sites) are sequenced at least four times with respect to a circular consensus sequence.

FIG. 130 shows MspI-based targeted single-molecule real-time sequencing without DNA end repair and A-tailing. In embodiments, the ligation between the digested DNA molecule and the hairpin adapter can be performed without the processes of DNA end repair and A-tailing. Digested DNA molecules with 5'CG overhangs can be directly ligated with hairpin adapters with 5'CG overhangs to form circular DNA templates for single-molecule real-time sequencing. After cleaning up unligated adapters and self-ligated adapter dimers, in some embodiments, unligated adapters, linear DNA, and incomplete circular DNA are removed prior to ligation with hairpin adapters. DNA molecules are suitable for single-molecule real-time sequencing, and IPD, PW, and sequence context can be obtained. The methylation profile of a single molecule will be determined using IPD, PW and sequence context according to this disclosure.

FIG. 131 shows MspI-based targeted single-molecule real-time sequencing with low likelihood of adapter self-ligation. Underlying cytosine bases refer to bases without a 5' phosphate group. In some embodiments, dephosphorylated hairpin adapters are used to minimize the possibility of self-ligated adapter dimer formation during the process of adaptor ligation, and these MspI-digested DNA molecules and adaptors. Concatenation can be done. These dephosphorylated hairpin adapters are unable to form self-ligated adapter dimers due to the lack of a 5' phosphate group. After ligation, the product is subjected to an adapter cleanup step to purify the hairpin adapter ligated DNA molecules. DNA molecules ligated with potentially nicked hairpin adapters are further subjected to phosphorylation (eg, T4 polynucleotide kinase) and nick sealing by DNA ligase (eg, T4 DNA ligase). In embodiments, removal of unligated adapters, linear DNA, and incomplete circular DNA can be further performed. DNA molecules ligated with hairpin adapters are suitable for IPD, PW, and single-molecule real-time sequencing to obtain sequence context. The methylation profile of a single molecule will be determined using IPD, PW and sequence context according to this disclosure.

In addition to MspI, other restriction enzymes such as SmaI, which contains the recognition site CCCGGG, can also be used.

In some embodiments, the desired size selection process can be performed after the DNA end repair step. In some embodiments, once the effect of a hairpin adapter on size selection results is determined, the desired size selection process can be performed after ligation of the hairpin adapter. In these and other embodiments, the order of the procedural steps involved in MspI-based targeted single-molecule real-time sequencing may vary depending on the experimental context.

In embodiments, size selection is performed using gel electrophoresis-based methods and/or magnetic bead-based methods. In embodiments, the restriction enzymes include, but are not limited to, BgIII, EcoRI, EcoRII, BamHI, HindIII, TaqI, NotI, HinFI, PvuII, Sau3AI, SmaI, HaeIII, HgaI, HpaII, AluI, EcoRV, EcoP15I, KpnI, PstI. , SacI, SalI, ScaI, SpeI, SphI, StuI, XbaI, and combinations thereof.

b) Distinguishing Biological Sample Types by Methylation This section describes how methylation profiles determined using fragments generated by restriction enzyme digestion can be used to facilitate discrimination between different biological samples. do.

Methylation profiles determined by MspI-based single-molecule real-time sequencing, according to embodiments of the present disclosure, were used to assess differences in methylation profiles between biological samples. Placental tissue DNA and buffy coat DNA samples were taken as an example. Computer simulations were performed to generate data for placental and buffy coat DNA samples based on MspI-based targeted single-molecule real-time sequencing. Simulations were based on kinetic values including IPD and PW for each nucleotide previously generated by SMRT sequencing placental tissue DNA and buffy coat DNA with whole-genome coverage using Sequel II Sequencing Kit 1.0. was Placental DNA and buffy coat DNA samples were then subjected to MspI digestion, after which a size range of 50-200 bp was used to simulate gel-based size selection conditions. Selected DNA molecules were ligated with hairpin adapters to form circular DNA templates. The circular DNA template was subjected to single-molecule real-time sequencing to obtain information on IPD, PW, and sequence context.

Assuming 500,000 ZMWs generating SMRT sequencing subreads, these subreads followed the genomic distribution of MspI digested fragments within the size range of 50-200 bp, as shown in Table 1. The sub-read depth was assumed to be 30-fold for both placental and buffy coat DNA samples. The simulation was repeated 10 times for each placental DNA sample and buffy coat DNA sample. Thus, the dataset generated in silico by MspI-digested targeted single-molecule real-time sequencing contained a total of 10 placental DNA samples and obtained 10 buffy coat DNA samples. The dataset was further analyzed by CNN to determine the methylation profile of each sample according to the present disclosure. CpG sites from a median of 9,198 CpG islands (range: 5,497-13,928) were obtained, and all sequenced CpG sites (range: 45,304-90,762) were obtained. It accounted for 13.6%. The methylation status of each CpG site in each molecule was determined by the CNN model according to this disclosure.

Figure 132 is a graph of global methylation levels between placental and buffy DNA samples determined by MspI-based targeted single-molecule real-time sequencing. The y-axis is the methylation level as a percentage. On the x-axis the sample type was listed. Figure 132 shows that overall methylation levels were higher in placenta samples (median: 57.6%) compared to buffy coat samples (median: 69.5%; range: 68.9%-70.4%). %; range: 56.9%-59.1%) (P value <0.0001, Mann-Whitney U test). These results suggested that methylation profiles determined by MspI-based single-molecule real-time sequencing could be used to distinguish between tissue or biological samples based on methylation differences. These data demonstrate that methylation differences detected by MspI-based single-molecule real-time sequencing can distinguish placental-derived DNA from buffy-coat DNA, suggesting that this method can be used in maternal plasma. can be applied to measure the fetal DNA fraction of Methylation is used to measure the fetal DNA fraction because the fetal DNA in maternal plasma or serum is derived from the placenta, while the remaining DNA molecules in the sample are primarily derived from maternal buffy coat cells. be able to. In embodiments, this technique is a useful tool for distinguishing between different tissues, or tissues with different diseases and/or physiological conditions, or biological samples.

To perform cluster analysis between placental and buffy coat DNA samples using the methylation profile of CpG islands, the percentage of CpG sites classified as methylated among all CpG sites of CpG islands was was used to calculate the DNA methylation levels of CpG islands. For illustrative purposes, cluster analysis was performed using methylation levels of CpG island regions.

Figure 133 shows cluster analysis of placenta and buffy coat samples using DNA methylation profiles determined by MspI-based targeted single-molecule real-time sequencing. The similarity of methylation patterns from CpG islands across different patients is indicated by the clustering dendrogram height values. In this example, height is calculated according to Euclidean distance. In one embodiment, a height cutoff of 100 can be used to divide the clustering tree into two groups and distinguish between placental and buffy coat samples with 100% sensitivity and specificity. Other height cutoffs may be used in other embodiments, including but not limited to 50, 60, 70, 80, 90, 120, 130, 140, and 150, and the like. FIG. 133 shows that 10 placental DNA samples and 10 buffy coat DNA samples were analyzed using CpG island methylation profiles determined by MspI-based single-molecule real-time sequencing according to the present disclosure. clearly clustered into one group.

V. Training and Detection Methods This section provides examples of how to train a machine learning model to detect base modifications and how to use a machine learning model to detect base modifications.

A. Model Training FIG. 102 illustrates an exemplary method 1020 for detecting modifications of nucleotides in nucleic acid molecules. An exemplary method 1020 can be a method of training a model to detect modifications. Modifications can include methylation. Methylation can include any methylation described herein. Modifications can have separate states, such as methylated and unmethylated, and may specify the type of methylation. Thus, a nucleotide may have more than two states (classifications).

At block 1022, a plurality of first data structures are received. Various examples of data structures are described herein, eg, in FIGS. 4-16. Each first data structure of the first plurality of first data structures may correspond to a respective window of sequenced nucleotides in a respective nucleic acid molecule of the plurality of first nucleic acid molecules. Each window associated with the first plurality of data structures may comprise 4 or more consecutive nucleotides, and 17, 18, 19, 20, 21 or more contiguous nucleotides are included. Each window may contain the same number of contiguous nucleotides. Windows may overlap. Each window can include nucleotides on the first strand of the first nucleic acid molecule and nucleotides on the second strand of the first nucleic acid molecule. The first data structure may also include a strand property value for each nucleotide in the window. Strand characteristics can indicate either the nucleotides present, or the first or second strand. The window may include nucleotides of the second strand that are not complementary to nucleotides at corresponding positions of the first strand. In some embodiments, all nucleotides on the second strand are complementary to nucleotides on the first strand. In some embodiments, each window may contain nucleotides from only one strand of the first nucleic acid molecule.

The first nucleic acid molecule can be a circular DNA molecule. A circular DNA molecule can be formed by cleaving a double-stranded DNA molecule, using the Cas9 complex to form the cleaved double-stranded DNA molecule. Hairpin adapters can be ligated to the ends of cleaved double-stranded DNA molecules. In embodiments, both ends of a double-stranded DNA molecule can be cleaved and ligated. For example, cleavage, ligation, and subsequent analysis may proceed as described in FIG.

the first plurality of first data structures are: 5,000-10,000; 10,000-50,000; 50,000-100,000; 100,000-200,000; 000, 500,000 to 1,000,000, or more than 1,000,000 first data structures. The plurality of first nucleic acid molecules comprises at least 1,000, 10,000, 50,000, 100,000, 500,000, 1,000,000, 5,000,000 or more nucleic acid molecules. obtain. As a further example, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 sequence reads can be generated.

Each of the first nucleic acid molecules is sequenced by measuring pulses of signals corresponding to nucleotides. The signal can be a fluorescent signal, or other type of optical signal (eg, chemiluminescence, photometry). A signal may result from a nucleotide or a tag attached to a nucleotide.

The modifications have a known first state of the nucleotide at the target position of each window of each first nucleic acid molecule. The first state can be that the modification is absent from the nucleotide or the modification is present at the nucleotide. The modification may be known to be absent from the first nucleic acid molecule, or the first nucleic acid molecule may be treated so that the modification is absent. The modification may be known to be present in the first nucleic acid molecule, or the first nucleic acid molecule may be treated such that the modification is present. If the first state is the absence of modifications, the modifications may not be present in each window of each first nucleic acid molecule, but only at the target position. The first known state may include a methylation state at a first location of the first data structure and an unmethylation state at a second location of the first data structure.

The target position can be the center of each window. For windows that span an even number of nucleotides, the target position can be the position immediately upstream or downstream of the center of the window. In some embodiments, the target position may be at any other position in each window, including the first position or the last position. For example, if the window spans n nucleotides on one strand, position 1 to position n (either upstream or downstream), then the target position can be any position from position 1 to position n. may be in position.

Each first data structure contains values for the properties within the window. A property can be for each nucleotide within the window. Properties can include nucleotide identities . An identity can include a base (eg, A, T, C, or G). The properties can also include the nucleotide position relative to the target position within each window. For example, a position can be a nucleotide distance to a target position. A position can be +1 if the nucleotide is one nucleotide away from the target position in one direction, and a position can be -1 if the nucleotide is one nucleotide away from the target position in the opposite direction.

A characteristic may include the width of a pulse corresponding to a nucleotide. The width of the pulse may be the width of the pulse at half its maximum value. The characteristics may further include a pulse interval (IPD) representing the time between pulses corresponding to a nucleotide and pulses corresponding to neighboring nucleotides. A pulse interval can be the time between a pulse maximum associated with a nucleotide and a pulse maximum associated with a neighboring nucleotide. Neighboring nucleotides can be adjacent nucleotides. The characteristics may also include the pulse height corresponding to each nucleotide within the window. The properties can further include a strand property value that indicates whether the nucleotide is on the first strand or the second strand of the first nucleic acid molecule. The representation of the strands can be similar to the matrix shown in FIG.

Each data structure of the plurality of first data structures may exclude first nucleic acid molecules having widths less than the IPD or cutoff value. Only first nucleic acid molecules that have an IPD value greater than the 10th percentile (or the 1st, 5th, 15th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, 90th, or 95th percentile) can be used. Percentiles may be based on data from the reference sample or all nucleic acid molecules within the reference sample. Width cutoff values may also correspond to percentiles.

At block 1024, a plurality of first training samples are stored. Each first training sample includes one of the first plurality of first data structures and a first label indicative of a first state of modification of the nucleotide at the target position.

At block 1026, a second plurality of second data structures is received. Block 1026 may be optional. Each second data structure of the second plurality of second data structures corresponds to a respective window of sequenced nucleotides in a respective nucleic acid molecule of the plurality of second nucleic acid molecules. The second plurality of nucleic acid molecules may be the same or different than the first plurality of nucleic acid molecules. Modifications have a known second state of the nucleotide at the target position within each window of each second nucleic acid molecule. The second state is a state different from the first state. For example, if there is a modification in the first state, there is no modification in the second state, and vice versa. Each second data structure contains values for the same property as the first plurality of first data structures.

A plurality of first training samples can be generated using multiple displacement amplification (MDA). In some embodiments, a plurality of first training samples can be generated by amplifying a first plurality of nucleic acid molecules using a set of nucleotides. A set of nucleotides may contain a first type of methylation (eg, 6mA or any other methylation [eg, CpG]) in a specific ratio. Specified ratios can include 1:10, 1:100, 1:1000, 1:10000, 1:100000, or 1:1000000 relative to unmethylated nucleotides. A plurality of second nucleic acid molecules can be generated using multiple substitution amplification with unmethylated nucleotides of the first type.

At block 1028, a plurality of second training samples are stored. Block 1028 may be optional. Each second training sample includes one of the second plurality of second data structures and a second label indicative of a second state of modification of the nucleotide at the target position.

At block 1029, the model is trained using a plurality of first training samples and optionally a plurality of second training samples. Training is performed on a first label and optionally a second label when a first plurality of first data structures and optionally a second plurality of second data structures are input to the model. This is done by optimizing the parameters of the model based on the model's outputs that match or do not match the corresponding labels. The output of the model specifies whether the nucleotide at the target position in each window has the modification. The method may only include a plurality of first training samples, as the model may identify outliers as different states than the first state. A model can be a statistical model, also called a machine learning model.

In some embodiments, the output of the model may include probabilities at each of multiple states. The state with the highest probability can be considered that state.

A model may include a convolutional neural network (CNN). A CNN may include a set of convolution filters configured to filter a first plurality of data structures and optionally a second plurality of data structures. The filter can be any filter described herein. The number of filters in each layer is 10-20, 20-30, 30-40, 40-50, 50-60, 60-70, 70-80, 80-90, 90-100, 100-150, 150-200 , or more. The filter kernel size can be 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15-20, 20-30, 30-40, or larger. could be. The CNN may include an input layer configured to receive a first plurality of filtered data structures and, optionally, a second plurality of filtered data structures. A CNN may also include multiple hidden layers containing multiple nodes. The input layer was combined with the first of the multiple hidden layers. The CNN may further include an output layer coupled to the last layer of the plurality of hidden layers and configured to output an output data structure. The output data structure may contain properties.

Models may include supervised learning models. Supervised learning models may include different approaches and algorithms, analytic learning, artificial neural networks, backpropagation, boosting (meta-algorithms), Bayesian statistics, case-based inference, decision tree learning, inductive logic. programming, Gaussian process regression, genetic programming, group methods of data processing, kernel estimators, learning automata, learning classifier systems, minimum message length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifiers , maximum entropy classifiers, conditional random fields, nearest neighbor algorithms, probabilistic and approximately correct learning (PAC) learning, ripple-down rules, knowledge acquisition methodologies, symbolic machine learning algorithms, sub-symbolic machine learning algorithms, support vector machines , Minimum Complexity Machines (MCM), Random Forests, Ensemble of Classifiers, Regular Classification, Data Pre-Processing, Processing of Imbalanced Data Sets, Statistical Relationship Learning, or Proaftn, Multi-Criteria Classification Algorithms. Models include linear regression, logistic regression, deep recurrent neural networks (e.g. long short-term memory, LSTM), Bayesian classifiers, hidden Markov models (HMM), linear discriminant analysis (LDA), k-means clustering, applications with noise density-based spatial clustering (DBSCAN), random forest algorithm, support vector machine (SVM), or any model described herein.

As part of training a machine learning model, the parameters of the machine learning model (weights, thresholds, etc., that can be used, for example, in the activation function of a neural network) are optimized based on a training sample (training set), Provides optimized accuracy in classifying nucleotide modifications at target positions. Various forms of optimization can be performed, such as backpropagation, empirical risk minimization, and structural risk minimization. A validation set of samples (data structures and labels) can be used to validate the accuracy of the model. Cross-validation can be performed using various portions of the training set for training and validation. A model can contain multiple sub-models, thereby providing an ensemble model. The sub-models can be weaker models, but when combined they provide a more accurate final model.

In some embodiments, chimeric or hybrid nucleic acid molecules can be used to validate models. At least some of the plurality of first nucleic acid molecules each include a first portion corresponding to the first reference sequence and a second portion corresponding to the second reference sequence. The first reference sequence can be from a different chromosome, tissue (eg, tumor or non-tumor), organism, or species than the second reference sequence. The first reference sequence can be human and the second reference sequence can be from a different animal. Each chimeric nucleic acid molecule can comprise a first portion corresponding to a first reference sequence and a second portion corresponding to a second reference sequence. The first location can have a first methylation pattern and the second location can have a second methylation pattern. The first site can be treated with a methylase. The second position may not be treated with a methylase and may correspond to an unmethylated position in the second reference sequence.

B. Detecting Modifications FIG. 103 shows a method 1030 for detecting modifications of nucleotides in a nucleic acid molecule. The modification can be any modification described in method 1020 of FIG.

At block 1032, an input data structure is received. The input data structure may correspond to a window of sequenced nucleotides in a sample nucleic acid molecule. A sample nucleic acid molecule can be sequenced by measuring pulses of light signals corresponding to nucleotides. The window can be any window described in block 1022 of FIG. 102 and the sequencing can be any sequence described in block 1022 of FIG. The input data structure may contain values for the same properties as described in block 1022 of FIG. Method 1030 can include sequencing the sample nucleic acid molecules.

Nucleotides within the window may or may not be aligned to the reference genome. Nucleotides within the window can be determined using circular consensus sequences (CCS) without aligning the sequenced nucleotides to a reference genome. The nucleotides in each window may be specified by CCS rather than aligned to the reference genome. In some embodiments, the window can be determined without using CCS and without aligning the sequenced nucleotides to a reference genome.

Nucleotides within the window can be enriched or filtered. Enrichment may be by an approach involving Cas9. The Cas9 approach uses the Cas9 complex to cleave a double-stranded DNA molecule to form a cleaved double-stranded DNA molecule, similar to FIG. can include ligating to the end of the Filtering may be by selecting double-stranded DNA molecules having a size within a size range. Nucleotides may be derived from these double-stranded DNA molecules. Other methods of maintaining the methylation status of molecules can be used (eg, methyl-binding proteins).

At block 1034, the input data structure is input to the model. The model may be trained by method 1020 of FIG.

In some embodiments, chimeric nucleic acid molecules can be used to validate models. At least some of the plurality of first nucleic acid molecules each have a first location corresponding to a first reference sequence and a second location corresponding to a second reference sequence different from the first reference sequence. including. The first reference sequence may be a different chromosome, tissue (e.g., tumor or non-tumor), organelle (e.g., mitochondria, nucleus, chloroplast), organism (mammal, virus, bacteria) than the second reference sequence. etc.), or from a species. The first reference sequence can be human and the second reference sequence can be from a different animal. Each chimeric nucleic acid molecule can comprise a first portion corresponding to a first reference sequence and a second portion corresponding to a second reference sequence. The first location can have a first methylation pattern and the second location can have a second methylation pattern. The first site can be treated with a methylase. The second position may not be treated with a methylase and may correspond to an unmethylated position in the second reference sequence.

At block 1036, it is determined using the model whether the modification is present at the nucleotide at the target position within the window of the input data structure.

The input data structure may be one of a plurality of input data structures. Each input data structure may correspond to a respective window of sequenced nucleotides in a respective sample nucleic acid molecule of the plurality of sample nucleic acid molecules. A plurality of sample nucleic acid molecules can be obtained from the subject's biological sample. The biological sample can be any biological sample described herein. Method 1030 can be repeated for each input data structure. The method may include receiving multiple input data structures. Multiple input data structures can be input to the model. A model can be used to determine whether a modification exists at the nucleotide at the target position within each window of each input data structure.

Each sample nucleic acid molecule of the plurality of sample nucleic acid molecules can have a size greater than the cutoff size. For example, cut-off sizes are 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 2 kb, 3 kb, 4 kb, 5 kb, 6 kb, 7 kb, 9 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb. , 60 kb, 70 kb, 80 kb, 90 kb, 100 kb, 500 kb, or 1 Mb. Having a size cutoff can lead to higher sub-read depths, and in both cases can increase the accuracy of modification detection. In some embodiments, the method may comprise fractionating the DNA molecules for a particular size prior to sequencing the DNA molecules.

Multiple sample nucleic acid molecules can be aligned to multiple genomic regions. For each genomic region of the plurality of genomic regions, several sample nucleic acid molecules can be aligned to the genomic region. The number of sample nucleic acid molecules may be greater than the cutoff number. The cutoff number may be a sub-read depth cutoff. The cutoff numbers for the sub-read depth are 1, 10, 30, 40, 50, 60, 70, 80, 900, 100, 200, 300, 400, 500, It can be 600-fold, 700-fold, or 800-fold. A sub-read depth cutoff number can be determined to improve or optimize accuracy. The sub-read depth cutoff number may be related to the number of multiple genomic regions. For example, the higher the sub-read depth cutoff number, the lower the number of multiple genomic regions.

Modifications can be determined to be present at one or more nucleotides. Disorder classification can be determined using the presence of one or more nucleotide modifications. Classification of disorders may involve using a number of modifications. The number of modifications can be compared to a threshold. Alternatively or additionally, a classification may include one or more positions of modification. The location of one or more modifications can be determined by aligning the sequence reads of the nucleic acid molecule to a reference genome. A disorder can be determined when a modification is shown to be at a particular position that is known to be correlated with the disorder. For example, the pattern of methylation sites can be compared to a reference pattern of disorders, and the disorder can be determined based on the comparison. A match to the reference pattern or a substantial match to the reference pattern (eg, 80%, 90%, or 95% or more) may indicate a disorder or a high probability of a disorder. The disorder can be cancer or any disorder described herein (eg, pregnancy-related disorders, autoimmune diseases).

A statistically significant number of nucleic acid molecules can be analyzed to accurately determine a lesion, tissue origin, or clinically relevant DNA fraction. In some embodiments, at least 1,000 nucleic acid molecules are analyzed. In other embodiments, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 or more nucleic acid molecules can be analyzed. As a further example, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 sequence reads can be generated.

The method may include classifying the disorder including determining that the subject has the disorder. Classification can include level of disorder using the number and/or site of modification.

Clinically relevant DNA fractions, fetal methylation profiles, maternal methylation profiles, presence of imprinted gene regions, or tissue of origin (e.g., from a sample containing a mixture of different cell types) can be identified by one or more nucleotide can be determined using the presence of modifications of Clinically relevant DNA fractions include, but are not limited to, fetal DNA fractions, tumor DNA fractions (e.g., from samples containing a mixture of tumor and non-tumor cells), and graft DNA fractions (e.g., donor from samples containing mixtures of cells and recipient cells).

The method may further include treating the disorder. Treatment can be provided according to the level of damage determined, the modification identified, and/or the tissue of origin (eg, of tumor cells isolated from the circulation of a cancer patient). For example, identified modifications can be targeted using specific drugs or chemotherapy. The tissue of origin can be used to guide surgery or any other form of treatment. Also, the level of injury can be used to determine how aggressive any type of treatment is.

Embodiments may include treating the disorder in the patient after determining the level of disorder in the patient. Treatment may include any suitable therapy, drug, chemotherapy, radiation, or surgery, including any treatment described in the references mentioned herein. The information regarding therapy in the references is incorporated herein by reference.

VI. Haplotype Analysis Differences in methylation profiles between the two haplotypes were found in tumor tissue samples. Therefore, methylation imbalance between haplotypes can be used to determine the level classification of cancer or other disorders. Haplotype imbalance can also be used to identify inheritance of haplotypes by the fetus. Fetal disorders can also be identified through analysis of methylation imbalance between haplotypes. Cellular DNA can be used to analyze haplotype methylation levels.

A. Haplotype-associated methylation analysis Single-molecule real-time sequencing technology allows individual SNPs to be identified. Long reads (e.g., up to several kilobases) generated from single-molecule real-time sequencing wells are capable of phasing genomic variants by exploiting the haplotype information present in each consensus read. (Edge et al. Genome Res. 2017;27:801-812, Wenger et al. Nat Biotechnol. 2019;37:1155-1162). Haplotype methylation profiles can be analyzed from the methylation levels of CpG sites linked to each haplotype allele by CCS, as shown in FIG. This stepwise methylation haplotype analysis can be used to resolve questions as to whether two copies of homologous chromosomes share similar or different methylation patterns in different clinically relevant conditions such as cancer. . In one embodiment, the methylation of a haplotype will be the aggregate methylation level contributed by several DNA fragments assigned to that haplotype. Haplotypes can be blocks of different sizes, including but not limited to 50 nt, 100 nt, 200 nt, 300 nt, 400 nt, 500 nt, 1 knt, 2 knt, 3 knt, 4 knt, 5 knt, 10 knt, 20 knt, 30 knt, 40 knt, 50 knt, 100 knt, 200 knt , 300 knt, 400 knt, 500 knt, 1 Mnt, 2 Mnt, and 3 Mnt.

B. Relative Haplotype-Based Methylation Imbalance Analysis FIG. 104 shows a relative haplotype-based methylation imbalance analysis. Haplotypes (ie, Hap I and Hap II) were determined by analyzing the results of single-molecule real-time sequencing. The methylation pattern linked to each haplotype can be determined using haplotype-associated fragments whose methylation profiles have been determined according to the approach described in FIG. It allows comparison of methylation patterns between Hap I and Hap II.

To quantify the methylation difference between Hap I and Hap II, the difference in methylation level (ΔF) between Hap I and Hap II was calculated. The difference ΔF is calculated as follows.
ΔF=M _HapI −M _HapII
where ΔF represents the difference in methylation level between Hap I and Hap II, and M _{Hap I} and M _Hap II represent the methylation levels of Hap I and Hap II, respectively. A positive value for ΔF suggests a higher DNA methylation level for Hap I compared to Hap II.

C. Relative Haplotype-Based Methylation Imbalance Analysis of HCC Tumor DNA In one embodiment, haplotype methylation analysis can be useful for detecting methylation abnormalities in cancer genomes. For example, methylation changes between two haplotypes within a genomic region are analyzed. Haplotypes within a genomic region are defined as haplotype blocks. A haplotype block can be viewed as a staged set of chromosomal alleles. In some embodiments, the haplotype block is extended as long as possible according to the set of sequence information supporting two alleles physically linked on the chromosome. For case 3033, 97,475 haplotype blocks were obtained from sequencing results of adjacent normal tissue DNA. The median haplotype block size was 2.8 kb. 25% of the haplotype blocks exceeded 8.2 kb in size. The maximum size of the haplotype block was 282.2 kb. The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

To illustrate, several criteria were used to identify potential haplotype blocks that showed differential methylation between Hap I and Hap II of tumor DNA compared to adjacent non-tumor tissue DNA. bottom. The criteria were as follows. (1) The haplotype blocks analyzed contained at least three triplicate CCS sequences each generated from triplicate sequencing wells. (2) The absolute difference in methylation levels between Hap I and Hap II in adjacent non-tumor tissue DNA was less than 5%. (3) The absolute difference in methylation levels between Hap I and Hap II in tumor tissue DNA exceeded 30%. We identified 73 haplotype blocks that met the above criteria.

Figures 105A and 105B are tables of 73 haplotype blocks showing differential methylation levels between Hap I and Hap II in HCC tumor DNA compared to adjacent non-tumor tissue DNA of case TBR3033. The first column indicates the chromosomes associated with the haplotype block. The second column indicates the starting coordinates of the haplotype block within the chromosome. The third column shows the ending coordinates of the haplotype block. The fourth column indicates the length of the haplotype block. The fourth column lists the ID of the haplotype block. The fifth column shows Hap I methylation levels in non-tumor tissue adjacent to tumor tissue. The sixth column shows Hap II methylation levels in non-tumor tissues. The seventh column shows Hap I methylation levels in tumor tissue. The eighth column shows Hap II methylation levels in tumor tissue.

In contrast to the 73 haplotype block showing greater than 30% difference in methylation levels between haplotypes in tumor tissue DNA, greater than 30% difference in non-tumor tissue DNA but less than 5% in tumor tissue DNA Only one haplotype block showed a difference in In some embodiments, another set of criteria can be used to identify haplotype blocks exhibiting differential methylation. Other maximum and minimum threshold differences can be used. For example, the minimum threshold difference can be 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, or more. As examples, the maximum threshold difference is 1%, 5%, 10%, 15%, 20%, or 30%. These results suggested that variations in methylation differences between haplotypes may serve as new biomarkers for cancer diagnosis, detection, surveillance, prognosis, and guidance for treatment.

In some embodiments, when studying methylation patterns, long haplotype blocks are split in silico into smaller blocks.

For case 3032, 61,958 haplotype blocks were obtained from sequencing results of adjacent non-tumor tissue DNA. The median haplotype block size was 9.3 kb. 25% of the haplotype blocks exceeded 27.6 kb in size. The maximum size of the haplotype block was 717.8 kb. As an example, the same three criteria as above were used to identify potential haplotype blocks that showed differential methylation between Hap I and Hap II of tumor DNA compared to adjacent normal tissue DNA. . Twenty haplotype blocks were identified that met the above criteria. The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

FIG. 106 is a table of 20 haplotype blocks showing differential methylation levels between Hap I and Hap II in tumor DNA compared to adjacent normal tissue DNA of case TBR3032. The first column indicates the chromosomes associated with the haplotype block. The second column indicates the starting coordinates of the haplotype block within the chromosome. The third column shows the ending coordinates of the haplotype block. The fourth column indicates the length of the haplotype block. The fourth column lists the ID of the haplotype block. The fifth column shows Hap I methylation levels in non-tumor tissue adjacent to tumor tissue. The sixth column shows Hap II methylation levels in non-tumor tissues. The seventh column shows Hap I methylation levels in tumor tissue. The eighth column shows Hap II methylation levels in tumor tissue.

In contrast to the 20 haplotype blocks showing differences in HCC tumor tissues in Figure 106, only one haplotype block showed greater than 30% difference in non-tumor tissue, but less than 5% difference in tumor tissue. showed that. These results further suggest that variations in methylation differences between haplotypes may serve as novel biomarkers for cancer diagnosis, detection, surveillance, prognosis, and guidance for treatment. ing. In other embodiments, other criteria can be used to identify haplotype blocks that exhibit differential methylation.

D. Relative Haplotype-Based Methylation Imbalance Analysis of DNA from Other Tumor Types As described above, analysis of methylation levels between haplotypes showed that HCC tumor tissues compared to paired adjacent non-tumor tissues were: We found that we had more haplotype blocks indicating methylation imbalance. As an example, the criteria for haplotype block showing methylation imbalance in tumor tissue were as follows. (1) The haplotype block analyzed contained at least three CCS sequences generated from three sequencing wells. (2) The absolute difference in methylation levels between Hap I and Hap II in adjacent non-tumor or normal tissue DNA based on historical data was less than 5%. (3) The absolute difference in methylation levels between Hap I and Hap II in tumor tissue DNA exceeded 30%. Criterion (2) was included because non-tumor/normal tissues showing haplotype imbalance at the methylation level may represent imprinted regions rather than tumor regions. Criteria for haplotype block indicating methylation imbalance in non-tumor tissues were as follows. (1) The haplotype block analyzed contained at least three CCS sequences generated from three sequencing wells. (2) The absolute difference in methylation levels between Hap I and Hap II in adjacent non-tumor or normal tissue DNA based on historical data exceeded 30%. (3) The absolute difference in methylation levels between Hap I and Hap II in tumor tissue DNA was less than 5%.

Other criteria may be used in other embodiments. For example, to identify an imbalanced haplotype I cancer genome, in non-tumor tissues, differences in methylation levels between Hap I and Hap II were 1%, 5%, 10%, 20%, 40% %, 50%, or less than 60%, etc., and in tumor tissue the difference in methylation levels between Hap I and Hap II is 1%, 5%, 10%, 20%, 40% , 50%, or greater than 60%, and the like. To identify imbalanced haplotype I non-cancer genomes, the difference in methylation levels between Hap I and Hap II was 1%, 5%, 10%, 20%, 40% in non-tumor tissues. , 50%, or more than 60%, etc., while in tumor tissue the difference in methylation levels between Hap I and Hap II can be 1%, 5%, 10%, 20%, 40%, etc. %, 50%, or less than 60%.

FIG. 107A is a table summarizing the number of haplotype blocks showing methylation imbalance between two haplotypes between tumor and adjacent non-tumor tissue, based on data generated by Sequel II Sequencing Kit 2.0. is. The first column lists the tissue type. The second column lists the number of haplotype blocks showing methylation imbalance between the two haplotypes in tumor tissue. The third column lists the number of haplotype blocks showing methylation imbalance between the two haplotypes in paired adjacent non-tumor tissues. These rows show that there are more haplotype blocks indicating methylation imbalance between the two haplotypes in tumor tissue than in paired adjacent non-tumor tissue.

The median length of the haplotype blocks included in this analysis was 15.7 kb (IQR: 10.3-26.1 kb). These data, including the liver HCC results, indicate that in seven tissue types, tumor tissue has more haplotype blocks with methylation imbalance. In addition to liver, other tissues include colon, breast, kidney, lung, prostate, and stomach tissue. Thus, in some embodiments, the number of haplotype blocks with methylation imbalance can be used to detect whether a patient has a tumor or cancer.

FIG. 107B is a table summarizing the number of haplotype blocks showing methylation imbalance between two haplotypes in tumor tissues of different tumor stages, based on data generated by Sequel II Sequencing Kit 2.0. The first column indicates histological types with tumors. The second column shows the number of haplotype blocks with methylation imbalance between the two haplotypes in tumor tissue. The third column lists tumor staging information using the TNM classification of malignancies. T3 and T3a are larger size tumors than T2.

The table shows that in both breast and kidney, the larger the tumor, the more haplotype blocks exhibiting methylation imbalance. For example, for breast tissue, tissue classified as tumor grade T3 (TNM staging), ER positive, and exhibiting ERBB2 amplification is tumor grade T2 (TNM staging), PR (progesterone receptor)/ER There were more haplotype blocks showing methylation imbalance (57) than tissues classified as (estrogen receptor) positive, no ERBB2 amplification (18). For kidney tissue, tissue classified as tumor grade T3a had more haplotype blocks (68) showing methylation imbalance than tissue classified as tumor grade T2 (0).

In some embodiments, haplotype blocks exhibiting methylation imbalance can be utilized for tumor classification and to correlate with their clinical behavior (e.g., progression, prognosis, or therapeutic response). can. These data suggested that the degree of haplotype-based methylation imbalance could serve as a tumor classifier and could be incorporated into clinical studies or trials or eventual clinical services. Tumor classification may include size and severity.

E. Haplotype-Based Methylation Analysis of Maternal Plasma Cell-Free DNA Haplotypes of both parents or either parent can be determined. Haplotyping methods include long-read single-molecule sequencing, linked short-read sequencing (eg, 10x genomics), long-range single-molecule PCR, or population inference. If the paternal haplotype is known, the cell-free fetal DNA methylome can be determined by linking the methylation profiles of multiple cell-free DNA molecules each containing at least one paternal-specific SNP allele present along the paternal haplotype. can be constructed. In other words, the paternal haplotype is used as a scaffold to link fetal-specific lead sequences.

Figure 108 shows analysis of haplotypes for relative methylation imbalance. If the maternal haplotype is known, the methylation imbalance between the two haplotypes (ie, Hap I and Hap II) can be used to determine the maternal haplotype inherited to the fetus. As shown in Figure 108, plasma DNA molecules from pregnant women are sequenced using single-molecule real-time sequencing technology. Methylation and allelic information can be determined according to the disclosures herein. In one embodiment, SNPs associated with disease-causing genes are assigned as Hap I. When the fetus inherits Hap I, fragments with the Hap I allele are more abundant in the maternal plasma compared to those with the Hap II allele. Hypomethylation of fetal-derived DNA fragments reduces the methylation level of Hap I relative to that of Hap II. As a result, if Hap I methylation exhibits a lower level of methylation than Hap II, the fetus is more likely to inherit maternal Hap I. Otherwise, the fetus is more likely to inherit maternal Hap II. Clinical trials have used haplotype-based methylation imbalance analysis to determine that fetuses inherit maternal haplotypes associated with genetic disorders such as, but not limited to, fragile X syndrome, muscular dystrophy, Huntington's disease or beta-thalassemia. can decide whether

F. Example Method for Classifying Disorders FIG. 109 illustrates an exemplary method 1090 for classifying disorders in organisms having a first haplotype and a second haplotype. Method 1090 includes comparing relative methylation levels between the two haplotypes.

At block 1091, DNA molecules from the biological sample are analyzed to identify their location in the reference genome corresponding to the organism. The DNA molecule can be a cellular DNA molecule. For example, a DNA molecule can be sequenced to obtain sequence reads, which can be mapped (aligned) to a reference genome. If the organism is human, the reference genome is potentially a reference human genome from a particular subpopulation. As another example, a DNA molecule can be analyzed (e.g., after PCR or other amplification) with different probes, each covering one or more CpG sites that are heterozygous, as described below. correspond to possible genomic locations.

Additionally, a DNA molecule can be analyzed to determine each allele of the DNA molecule. For example, the allele of a DNA molecule can be determined from sequence reads obtained from sequencing or from specific probes that hybridize to the DNA molecule, both techniques can provide sequence reads (e.g. , the probes can be treated as sequence reads when hybridized). For a DNA molecule, the methylation status at each of one or more sites (eg, CpG sites) can be determined.

At block 1092, one or more heterozygous loci at the first location of the first chromosomal region are identified. Each heterozygous locus can comprise a corresponding first allele of the first haplotype and a corresponding second allele of the second haplotype. The one or more heterozygous loci may be the first plurality of heterozygous loci and the second plurality of heterozygous loci may correspond to different chromosomal regions.

At block 1093, a first set of multiple DNA molecules is identified. Each of the plurality of DNA molecules is located at any one of the heterozygous loci from block 1096 and contains the corresponding first allele, so that the DNA molecule is regarded as corresponding to the first haplotype can be specified. A DNA molecule can be located at more than one heterozygous locus, but typically a read contains only one heterozygous locus. Each of the first set of DNA molecules also includes at least one of the N genomic sites, and the genomic sites are used to measure methylation levels. N is an integer, such as 1, 2, 3, 4, 5, 10, 20, 50, 100, 200, 500, 1,000, 2,000, or 5,000 or greater. Thus, a DNA molecule read can exhibit 1-site, 2-site, etc. coverage. A genomic site can include the site where a CpG nucleotide is present.

At block 1094, a first methylation level at the first location of the first haplotype is determined using the first set of multiple DNA molecules. The first methylation level can be determined by any method described herein. The first location may correspond to a single site or may include many sites. The first location of the first haplotype can be 1 kb or greater. For example, the first location of the first haplotype may be 1 kb, 5 kb, 10 kb, 15 kb, or 20 kb or more. Methylation data may be data from cellular DNA.

In some embodiments, multiple first methylation levels may be determined for multiple locations of the first haplotype. Each locus can have a chain length of 5 kb or longer, or any size disclosed herein for the first locus of the first haplotype.

At block 1095, a second set of multiple DNA molecules is identified. Each of the plurality of DNA molecules is located at any one of the heterozygous loci from block 1096 and contains the corresponding second allele, so that the DNA molecule is considered to correspond to the second haplotype. can be specified. Each of the second set of DNA molecules also includes at least one of the N genomic sites, and the genomic sites are used to measure methylation levels.

At block 1096, a second methylation level at the first location of the second haplotype is determined using a second set of a plurality of DNA molecules. It can be determined by any method described. The first portion of the second haplotype may be greater than 1 kb or longer than any size of the first portion of the first haplotype. The first position of the first haplotype can be complementary to the first position of the second haplotype. The first portion of the first haplotype and the first portion of the second haplotype can form a circular DNA molecule. A first methylation level at the first location of the first haplotype can be determined using data from the circular DNA molecule. For example, analysis of circular DNA can include the analysis described in FIGS. 1, 2, 4, 5, 6, 7, 8, 50, or 61.

A circular DNA molecule can be formed by cleaving a double-stranded DNA molecule, using the Cas9 complex to form the cleaved double-stranded DNA molecule. Hairpin adapters can be ligated to the ends of cleaved double-stranded DNA molecules. In embodiments, both ends of a double-stranded DNA molecule can be cleaved and ligated. For example, cleavage, ligation, and subsequent analysis may proceed as described in FIG.

In some embodiments, multiple second methylation levels may be determined for multiple locations of the second haplotype. Each location of the plurality of locations of the second haplotype can be complementary to a location of the plurality of locations of the first haplotype.

At block 1097, the value of the parameter is calculated using the first methylation level and the second methylation level. This parameter may depend on the separation value. A separation value may be the difference between two methylation levels or the ratio of two methylation levels.

When multiple locations of the second haplotype are used, for each location of the multiple locations of the second haplotype, the separation value is the second methylation level of the location of the second haplotype and the It can be calculated using the first methylation level using complementary points. A separation value can be compared to a cutoff value.

A cut-off value can be determined from non-disturbed tissue. The parameter may be the number of locations of the second haplotype where the separation value exceeds the cutoff value. For example, the number of locations of the second haplotype whose segregation value exceeds the cutoff value is similar to the number of regions shown to have greater than 30% difference in Figures 105A, 105B, and 106. obtain. In Figures 105A, 105B, and 106, the cutoff value is a ratio and the cutoff value is 30%. In some embodiments, the cutoff value can be determined from tissue with the disorder.

In another example, the separation values for each location can be aggregated (eg, summed), which can be done by a weighted sum or functional sum of the respective separation values. Such aggregation can provide the value of the parameter.

At block 1098, the value of the parameter is compared with the reference value. Reference values can be determined using non-disturbed reference tissue. The reference value may be a separate value. For example, a reference value may represent that there should be no significant difference between the methylation levels of two haplotypes. For example, the reference value can be a statistical difference of 0 or a ratio of about 1. If multiple locations are used, the reference value can be the number of locations in healthy organisms where the two haplotypes show a separation value above the cutoff value. In some embodiments, the reference value can be determined using a reference tissue with a lesion.

At block 1099, a classification of the disorder in the organism is determined using a comparison of the parameter's value and the reference value. If the value of the parameter exceeds the reference value, it may be determined that a fault exists or is more likely. Disorders can include cancer. The cancer can be any cancer described herein. A classification of the disorder may be the likelihood of the disorder. Classification of disorders may include severity of disorders. For example, a larger parameter value indicating a higher number of sites with haplotype imbalance may indicate a more severe form of cancer.

Although the method described in FIG. 109 involves classification of disorders, similar methods can be used to determine any condition or trait that can result from imbalance in methylation levels between haplotypes. For example, haplotype methylation levels from fetal DNA may be lower than haplotype methylation levels from maternal DNA. Methylation levels can be used to classify nucleic acids as maternal or fetal.

If the disorder is cancer, different chromosomal regions of the tumor may exhibit such differences in methylation. Different treatments may be provided depending on the area affected. Furthermore, subjects with different regions exhibiting such differences in methylation may have different prognoses.

Chromosomal regions (locations) with sufficient segregation (eg, greater than a cutoff value) can be identified as abnormal (or with aberrant segregation). Patterns of abnormal regions (explaining which haplotypes are more likely than others) can be compared to reference patterns (e.g., subjects with cancer, potentially certain types of cancer, or healthy individuals). determined from the subject). If the two patterns are the same within a threshold than the reference pattern with a particular classification (e.g., less than a specified number of different regions/locations), the subject may be identified as having that classification for the disorder. Such classifications can include, for example, imprint disorders, as described herein.

VII. Single Molecule Methylation Analysis of Hybrid Molecules To further evaluate the performance and utility of the embodiments disclosed herein for determining base modifications of nucleic acids, the human portion was methylated and the mouse portion was unmethylated. We have engineered hybrid human and murine DNA fragments that have been modified and vice versa. By determining the junctions of hybrid or chimeric DNA molecules, it may be possible to detect gene fusions in various disorders or diseases, including cancer.

A. Methods for Making Hybrid Human and Mouse DNA Fragments This section describes procedures for making hybrid DNA fragments and then determining the methylation profile of the fragments.

In one embodiment, human DNA is amplified by whole genome amplification such that whole genome amplification does not preserve the methylation state, thereby eliminating the original methylation signature of the human genome. Whole-genome amplification can be performed using exonuclease-resistant thiophosphate-modified degenerate hexamers as primers that can bind randomly on the genome, allowing polymerases (e.g., Phi29 DNA polymerase) to amplify DNA without thermal cycling. it becomes possible to Amplified DNA products are unmethylated. Amplified human DNA molecules undergo the CpG methyltransferase M. Further treated with SssI. This theoretically fully methylates all cytosines in the CpG context in double-stranded, unmethylated, or hemimethylated DNA. Therefore, M. Such amplified human DNA treated with Sssl results in methylated DNA molecules.

In contrast, mouse DNA was subjected to whole genome amplification such that unmethylated mouse DNA fragments were generated.

Figure 110 shows the construction of a human-mouse hybrid DNA fragment in which the mouse portion is unmethylated and the human portion is methylated. Filled lollipops represent methylated CpG sites. Unfilled lollipops represent unmethylated CpG sites. The diagonally striped thick bars 11010 represent methylated human moieties. Vertically striped thick bars 11020 represent the unmethylated mouse portion.

For the generation of hybrid human-mouse DNA molecules, in one embodiment, whole genome amplification and M . The SssI-treated DNA molecule was further digested with HindIII and NcoI to generate sticky ends to facilitate downstream ligation. In one embodiment, methylated human DNA fragments were further mixed with unmethylated mouse DNA fragments in equimolar ratios. Such human-mouse DNA mixtures, in one embodiment, were subjected to a DNA ligase-mediated ligation process at 20° C. for 15 minutes. As shown in Figure 110, this ligation results in a human-mouse hybrid DNA molecule (a: human-mouse hybrid fragment), a human-only DNA molecule (b: human-human ligation, and c: unligated human Three types of resulting molecules are produced, including mouse-only DNA molecules (d: mouse-mouse ligation, and e: unligated mouse DNA). Post-ligation DNA products were subjected to single-molecule real-time sequencing. Sequencing results were analyzed according to the disclosure provided herein to determine methylation status.

Figure 111 shows the construction of a human-mouse hybrid DNA fragment in which the human portion is unmethylated and the mouse portion is methylated. Filled lollipops represent methylated CpG sites. Unfilled lollipops represent unmethylated CpG sites. The diagonally striped thick bar 11110 represents the methylated mouse portion. Vertically striped thick bars 11120 represent unmethylated human portions.

In the embodiment of Figure 111, mouse DNA molecules were amplified via whole genome amplification such that the original methylation of the mouse genome was eliminated. Amplified DNA products are unmethylated. Amplified mouse DNA was further isolated from M. processed with SssI. Therefore, M. Such amplified mouse DNA treated with Sssl results in methylated DNA molecules. In contrast, human DNA fragments were subjected to whole genome amplification so as to obtain unmethylated human fragments. In one embodiment, the methylated human fragment was further mixed with the unmethylated fragment in an equimolar ratio. Such human-mouse DNA mixtures were subjected to a ligation process mediated by DNA ligase. As shown in Figure 111, this ligation reaction results in a human-mouse hybrid DNA molecule (a: human-mouse hybrid fragment), a human-only DNA molecule (b: human-human ligation, and c: unligated human DNA). ), and mouse-only DNA molecules (d: mouse-mouse ligation, and e: unligated mouse DNA). Post-ligation DNA products were subjected to single-molecule real-time sequencing. Sequencing results were analyzed according to the disclosure provided herein to determine methylation status.

According to the embodiment shown in FIG. 110, the inventors prepared an artificial DNA mixture (designated sample MIX01), containing human-mouse hybrid DNA molecules, human-only DNA, and mouse-only DNA. , the human-associated DNA molecule was methylated and the mouse DNA molecule was unmethylated. For sample MIX01, 166 million subreads were obtained that could be aligned to either the human or mouse reference genome, or the partially human and partially mouse genome. These subreads were generated from approximately 5 million Pacific Biosciences single molecule real-time (SMRT) sequencing wells. Each molecule in a single-molecule real-time sequencing well was sequenced an average of 32 times (range: 1-881 times).

To determine the human and mouse DNA portions of hybrid fragments, a consensus sequence was first constructed by combining nucleotide information from all relevant subreads within a well. In total, 3,435,657 consensus sequences were obtained for sample MIX01. The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

Consensus sequences were aligned to reference genomes, including both human and mouse references. 3.2 million aligned consensus sequences were obtained. Among them, 39.6% of them were classified as human-only DNA type, 26.5% of them were classified as mouse-only DNA type, and 30.2% of them were human-mouse classified as hybrid DNA.

FIG. 112 shows the chain length distribution of DNA molecules in the DNA mixture after ligation (sample MIX01). The x-axis indicates the chain length of the DNA molecule. The y-axis indicates the frequency related to the length of the DNA molecule. As shown in Figure 112, the human-mouse hybrid DNA molecules had a longer chain length distribution, consistent with the fact that they are a combination of at least two types of molecules.

FIG. 113 shows junction regions where a first DNA (A) and a second DNA (B) are joined together. DNA (A) and DNA (B) can be digested with restriction enzymes. In one embodiment, to improve the efficiency of ligations using sticky ends, the restriction enzymes HindIII and NcoI (recognizing A^AGCTT and C^CATGG sites, respectively) were used prior to the ligation step to Human and mouse DNA were digested. DNA (A) and DNA (B) can then be ligated. Among 698,492 human-mouse hybrid DNA molecules with junction regions, 88% of human-mouse hybrid DNA molecules with enzyme recognition sites of A^AGCTT and C^CATGG were found, and human and mouse DNA suggesting that ligation between fragments occurred. The junction region is defined as the region or site where the first DNA segment and the second DNA segment are physically joined together. Since the junction contains sequences common to both DNA(A) and DNA(B), the point on one strand corresponding to the junction is either DNA(A) or DNA(B) by sequence alone. cannot be determined to be part of Analyzing the methylation pattern or density of a point on one strand corresponding to the junction can be used to determine whether the point is from DNA(A) or DNA(B). As an example, DNA (A) can be viral DNA and DNA (B) can be human DNA. Determination of precise junctions can inform whether and how such integrated DNA disrupts protein structure.

Figure 114 shows the methylation analysis of the DNA mixture. Diagonally striped bars 11410 indicate junctional regions observed in alignment analysis that would be introduced by restriction enzyme treatment prior to ligation. "RE site" refers to a restriction enzyme (RE) recognition site.

As shown in Figure 114, in one embodiment, the aligned consensus sequences were grouped into three categories as follows.

(1) Sequenced DNA was aligned only to the human reference genome and not to the mouse reference genome, with reference to one or more alignment criteria. In one embodiment, one alignment criterion includes, but is not limited to, 100%, 95%, 90%, 80%, 70%, 60%, 50%, 40%, 30% of the contiguous nucleotides of the sequenced DNA. %, or 20% can be aligned to the human reference. In one embodiment, one alignment criterion is that the remainder of the sequenced fragments that did not align to the human reference cannot align to the mouse reference genome. In one embodiment, one alignment criterion was that the sequenced DNA could be aligned to a single region of the reference human genome. In one embodiment, the alignment can be perfect. In still other embodiments, the alignment can accommodate nucleotide discrepancies, including insertions, mismatches, and deletions, provided that such discrepancies are below a certain threshold, including, but not limited to, the aligned sequences. such as 1%, 2%, 3%, 4%, 5%, 10%, 20%, or 30% of the length of the In another embodiment, the aligned can be at more than one location in the reference genome. In still other embodiments, the alignment to one or more sites of the reference genome is described in a probabilistic fashion (e.g., indicating possible misalignments) and the probability measure is used in subsequent processing. can be

(2) the sequenced DNA was aligned only to the mouse reference genome, but not to the human reference genome, with reference to one or more alignment criteria; In one embodiment, one alignment criterion includes, but is not limited to, 100%, 95%, 90%, 80%, 70%, 60%, 50%, 40%, 30% of the contiguous nucleotides of the sequenced DNA. %, or 20% can be defined as being aligned to the mouse reference. In one embodiment, one alignment criterion is that the remainder cannot be aligned to the human reference genome. In one embodiment, one alignment criterion was that the sequenced DNA could be aligned to a single region of the reference mouse genome. In one embodiment, the alignment can be perfect. In still other embodiments, the alignment can accommodate nucleotide discrepancies, including insertions, mismatches, and deletions, provided that such discrepancies are below a certain threshold, including, but not limited to, the aligned sequences. such as 1%, 2%, 3%, 4%, 5%, 10%, 20%, or 30% of the length of the In another embodiment, the aligned can be at more than one location in the reference genome. In still other embodiments, the alignment to one or more sites of the reference genome is described in a probabilistic fashion (e.g., indicating possible misalignments) and the probability measure is used in subsequent processing. can be

(3) A portion of the sequenced DNA was uniquely aligned with the human reference genome, while another portion was uniquely aligned with the mouse reference genome. In one embodiment, if restriction enzymes were used prior to ligation, alignment analysis would observe junction regions corresponding to restriction enzyme cleavage sites. In some embodiments, the junction region between the human and mouse DNA segments could only be approximately determined within certain regions due to sequencing and alignment errors. In some embodiments, at the junction region of the human-mouse hybrid DNA fragment, the restriction enzyme recognition site is not observed.

The pulse interval (IPD), pulse width (PW), and sequence context surrounding the CpG sites were obtained from those subreads corresponding to the consensus sequences. Thereby, the methylation of each DNA molecule, including human-only DNA, mouse-only DNA, and human-mouse hybrid DNA, could be determined according to the embodiments present in the present disclosure.

B. Methylation Results This section describes the results of methylation of hybrid DNA fragments. Methylation density can be used to identify the origin of various parts of the hybrid DNA fragment.

Figure 115 shows a boxplot of the probability that CpG sites in sample MIX01 are methylated. The x-axis indicates the three different molecules present in sample MIX01: human-only DNA, mouse-only DNA, and human-mouse hybrid DNA (containing both human and mouse portions). The y-axis indicates the probability that a particular single DNA molecule CpG site is methylated. This assay was performed in such a way that human DNA was more methylated and mouse DNA was more unmethylated.

As shown in FIG. 115, the probability that the CpG site is methylated in human-only DNA (median: 0.66, range: 0-1) is higher than that in mouse-only DNA (median: 0.66, range: 0-1). 06, range: 0-1) (P-value < 0.0001). These results were consistent with the assay design. Thus, human DNA contains the CpG methyltransferase M. Although it was more methylated due to SssI treatment, mouse DNA was more unmethylated due to failure to maintain methylation during whole genome amplification. Furthermore, the CpG sites within the human DNA portion of the human-mouse hybrid DNA molecule (median: 0.06, range: 0-1) are significantly less than the CpG sites within the mouse DNA portion (median: 0.69, range: 0). 1) was more likely to be methylated (P-value <0.0001). These data demonstrate that the disclosed method can accurately determine the methylation status of DNA molecules as well as segments within DNA molecules.

Methylation probability refers to the estimated probability of a particular CpG site within a single molecule based on the statistical model used. A probability of 1 indicates that 100% of the CpG sites are methylated, based on statistical models and using the measured parameters (including IPD, PW, and sequence context). A probability of 0 indicates that 0% of the CpG sites are methylated, based on statistical models and using the measured parameters (including IPD, PW, and sequence context). In other words, all CpG sites are unmethylated using the measured parameters. FIG. 115 shows the distribution of methylation probabilities, where the human-only DNA distribution and the human partial distribution are broader than their mouse counterparts. Bisulfite sequencing is used to measure methylation in similar samples to ensure that methylation is not complete. The results are shown below. Figure 115 shows significant differences between methylation in human versus mouse DNA.

According to the embodiment shown in FIG. 111, the inventors prepared an artificial DNA mixture (designated sample MIX02), containing human-mouse hybrid DNA molecules, human-only DNA, and mouse-only DNA. , the human part was unmethylated and the mouse part was methylated. For sample MIX02, 140 million subreads were obtained that could be aligned to either the human or mouse reference genome, or the partially human and partially mouse genome. These subreads were generated from approximately 5 million Pacific Biosciences single molecule real-time (SMRT) sequencing wells. Each molecule in a single-molecule real-time sequencing well was sequenced an average of 27 times (range: 1-1028 times).

We also constructed a consensus sequence by combining nucleotide information from all relevant subreads within a well. In total, 3,265,487 consensus sequences were obtained for sample MIX02. This consensus sequence was aligned using BWA to a reference genome containing both human and mouse references (Li H et al., Bioinformatics. 2010;26(5):589-595). Three million aligned consensus sequences were obtained. Of those, 30.5% were typed as human-only DNA type, 32.2% were typed as mouse-only DNA type, and 33.8% were typed as human-mouse hybrid DNA type. . The dataset was generated from DNA prepared by Sequel II Sequencing Kit 1.0.

Figure 116 shows the chain length distribution of DNA molecules in the DNA mixture after cross-ligation of sample MIX02. The x-axis indicates the chain length of the DNA molecule. The y-axis indicates the frequency related to the length of the DNA molecule. As shown in Figure 116, the human-mouse hybrid DNA molecules had a longer chain length distribution, consistent with the fact that they were produced by the ligation of two or more molecules.

FIG. 117 shows a boxplot of the probability that CpG sites are methylated in sample MIX02. Methylation status was determined according to the methods described herein. The x-axis indicates the three different molecules present in sample MIX01: human-only DNA, mouse-only DNA, and human-mouse hybrid DNA (containing both human and mouse portions). The y-axis shows the probability that the CpG site is methylated. This assay was performed in such a way that human DNA was unmethylated and mouse DNA was methylated.

As shown in FIG. 117, the probability of being methylated at a CpG site in human-only DNA (median: 0.06, range: 0-1) is higher than that in mouse-only DNA (median: 0.93). range: 0-1) was significantly lower (P value <0.0001). These results were consistent with the assay design. Thus, human DNA was more unmethylated due to the inability to maintain methylation during whole genome amplification, whereas in mouse DNA the CpG methyltransferase M. It was more methylated due to SssI treatment. Furthermore, the CpG sites within the human DNA portion of the human-mouse hybrid DNA molecule (median: 0.93, range: 0-1) are significantly less than the CpG sites within the mouse DNA portion (median: 0.07, range: 0). 1) was less likely to be methylated (P-value <0.0001). These data demonstrate that the disclosed method can accurately determine the methylation status of DNA molecules as well as segments within DNA molecules.

Bisulfite sequencing was used to measure methylation of human-mouse hybrid fragments whose methylation patterns were determined by single-molecule real-time sequencing according to embodiments of the present disclosure. Samples MIX01 (human DNA methylated, mouse DNA unmethylated) and sample MIX02 (human DNA unmethylated, mouse DNA methylated) were sheared via sonication and the central A mixture of DNA fragment sizes with a value of 196 bp was obtained (interquartile range: 161-268). Paired-end bisulfite sequencing (BS-Seq) was then performed using the MiSeq platform (Illumina) with a read length of 300 bp×2. 3.7 million and 2.9 million sequence fragments were obtained for MIX01 and MIX02, respectively, and aligned with human or mouse reference genomes, or partially human and partially mouse genomes. For MIX01, 41.6% of the aligned fragments were classified as human-only DNA, 56.6% as mouse-only DNA, and 1.8% as human-mouse hybrid DNA. For MIX02, 61.8% of the aligned fragments were classified as human-only DNA, 36.3% as mouse-only DNA, and 1.9% as human-mouse hybrid DNA. The percentage of sequenced fragments determined to be human-mouse hybrid DNA with BS-Seq (<2%) was much lower than the percentage observed with Pacific Biosciences sequencing results (>30%). In particular, the long fragments (median ~2 kb) were sequenced by Pacific Biosciences sequencing, but the long fragments were shared with the short fragments (median ~196 bp) suitable for MiSeq. Such a shearing process greatly dilutes the human-mouse hybrid fragment.

Figure 118 shows a table comparing methylation determined by bisulfite sequencing and Pacific Biosciences sequencing of MIX01. The leftmost section of the table indicates the type of DNA: 1) human-only, 2) mouse-only, and 3) human-mouse hybrids (divided into human and mouse portions). The middle section of the table shows details from the bisulfite sequencing, including the number of CG sites and methylation density. The rightmost section of the table shows details from Pacific Biosciences sequencing, including the number of CG sites and methylation density.

As shown in Figure 118, both bisulfite sequencing and Pacific Biosciences sequencing results showed that MIX01 human-only DNA consistently showed higher methylation densities than mouse-only DNA. For the human-mouse hybrid fragment, the bisulfite sequencing results determined the methylation levels of the human and mouse portions to be 46.8% and 2.3%, respectively. These results confirmed that the methylation density was higher in the human portion compared to the mouse portion, as determined by Pacific Biosciences sequencing according to the present disclosure. Pacific Biosciences sequencing observed a methylation density of 57.4% in the human part and a lower methylation density of 12.1% in the mouse part. These results suggest that methylation determined by Pacific Biosciences sequencing according to the present disclosure may be viable. In particular, Pacific Biosciences sequencing can be used to determine different methylation densities, including DNA that has a section with a higher methylation density than another section. Methylation densities determined by Pacific Biosciences sequencing according to the present disclosure were observed to be higher compared to bisulfite sequencing. Adjusting such estimates using the difference between results determined by these two techniques allows results to be compared across techniques.

Figure 119 shows a table comparing methylation determined by bisulfite sequencing and Pacific Biosciences sequencing of MIX02. The leftmost section of the table indicates the type of DNA: 1) human-only, 2) mouse-only, and 3) human-mouse hybrids (divided into human and mouse portions). The middle section of the table shows details from the bisulfite sequencing, including the number of CG sites and methylation density. The rightmost section of the table shows details from Pacific Biosciences sequencing, including the number of CG sites and methylation density.

As shown in Figure 119, both the bisulfite sequencing and Pacific Biosciences sequencing results showed that the human-only DNA of MIX02 consistently showed lower methylation densities than the mouse-only DNA. For the human-mouse hybrid fragment, the bisulfite sequencing results determined the methylation levels of the human and mouse portions to be 1.8% and 67.4%, respectively. These results further confirmed that the methylation density was lower in the human portion compared to the mouse portion, as determined by Pacific Biosciences sequencing according to the present disclosure. Pacific Biosciences sequencing observed a methylation density of 13.1% in the human portion and a higher methylation density of 72.2% in the mouse portion, as determined by Pacific Biosciences sequencing according to the present disclosure. was done. It also suggested that it would be feasible to determine methylation by Pacific Biosciences sequencing according to the present disclosure. In particular, Pacific Biosciences sequencing can be used to determine different methylation densities, including DNA having sections with lower methylation densities than other sections. It was also observed that the methylation density determined by Pacific Biosciences sequencing according to the present disclosure was higher compared to bisulfite sequencing. Adjusting such estimates using the difference between results determined by these two techniques allows results to be compared across techniques.

FIG. 120A shows methylation levels in 5 Mb bins of human-only and mouse-only DNA for MIX01. FIG. 120B shows methylation levels in 5 Mb bins of human-only and mouse-only DNA for MIX02. In both figures, the methylation level is shown in percent on the y-axis. Bisulfite sequencing and Pacific Biosciences sequencing for human-only DNA and mouse-only DNA, respectively, are shown on the x-axis.

In Figures 120A and 120B, across both bins of samples MIX01 and MIX02, results determined by Pacific Biosciences sequencing according to the present disclosure were found to be overall high.

FIG. 121A shows methylation levels in 5 Mb bins of the human and mouse portions of the human-mouse hybrid DNA fragment for MIX01. FIG. 121B shows methylation levels in 5 Mb bins of the human and mouse portions of the human-mouse hybrid DNA fragments for MIX02. In both figures, the methylation level is shown in percent on the y-axis. Bisulfite sequencing and Pacific Biosciences sequencing for the human and mouse portions of DNA, respectively, are shown on the x-axis.

Both Figures 121A and 121B showed an increase in methylation levels when using Pacific Biosciences sequencing compared to bisulfite sequencing. This increase is similar to the increase in methylation levels by Pacific Biosciences sequencing seen in human-only and mouse-only DNA in Figures 120A and 120B. The increased variability in methylation levels across the 5 Mb bins present in the bisulfite sequencing results of hybrid fragments was likely due to the low number of CpG sites used in the analysis.

Figures 122A and 122B are representative graphs showing the methylation status in a single human-mouse hybrid molecule. Figure 122A shows the human-mouse hybrid fragment within sample MIX01. Figure 122B shows the human-mouse hybrid fragment within sample MIX02. Filled circles indicate methylated sites and unfilled circles indicate unmethylated sites. The methylation status of these fragments was determined according to embodiments described herein.

As shown in Figure 122A, the human portion of the hybrid molecule from sample MIX01 was determined to be more methylated. In contrast, portions of mouse DNA were determined to be more hypomethylated. In contrast, Figure 122B shows that the human portion of the hybrid molecule from sample MIX02 was determined to be more hypomethylated, whereas the mouse DNA portion was determined to be more methylated. .

These results demonstrated that the embodiments present in the present disclosure allowed methylation changes to be determined in a single DNA molecule with different methylation patterns in different parts of the molecule. In one embodiment, the methylation status of a gene or other genomic region (eg, promoter versus gene body) can be measured where different portions of the gene or genomic region will exhibit different methylation states. In another embodiment, the methods presented herein are capable of detecting human-mouse hybrid fragments and detect DNA molecules containing non-contiguous fragments (i.e., chimeric molecules) with respect to the reference genome. , providing a general approach for analyzing their methylation status. For example, this approach can be used to analyze gene fusions, genome rearrangements, translation, inversions, duplications, structural changes, viral DNA integration, meiotic recombination, and the like, without limitation.

In some embodiments, these hybrid fragments may be enriched prior to sequencing using probe-based hybridization methods or CRISPR-Cas systems or their variant approaches for target DNA enrichment. Recently, it was reported that a CRISPR-associated transposase from the cyanobacterium Scytonema hofmanni can insert DNA segments into regions near the target site of interest (Strecker et al. Science. 2019;365:48-53). CRISPR-related transposases may function like Tn7-mediated transposition. In one embodiment, we can adapt this CRISPR-associated transposase to insert, for example, a biotin-labeled comment sequence into one or more gRNA-induced genomic regions of interest. can. For example, streptavidin-coated magnetic beads can be used to capture comment sequences, thereby simultaneously pulling down target DNA sequences for sequencing and methylation analysis according to embodiments of the present disclosure.

In some embodiments, fragments can be enriched by using restriction enzymes, which can include any of the restriction enzymes disclosed herein.

C. Examples of Methods for Detecting Chimeric Molecules FIG. 123 illustrates a method 1230 for detecting chimeric molecules in a biological sample. Chimeric molecules can comprise sequences from two different genes, chromosomes, organelles (eg, mitochondria, nucleus, chloroplast), organisms (mammals, bacteria, viruses, etc.), and/or species. Method 1230 can be applied to each of a plurality of DNA molecules from a biological sample. In some embodiments, the plurality of DNA molecules can be cellular DNA. In other embodiments, the plurality of DNA molecules can be cell-free DNA molecules from the plasma of pregnant women.

At block 1232, single molecule sequencing of the DNA molecule can be performed to obtain sequence reads that provide the methylation status at each of the N sites. N is 5 or greater, including 5-10, 10-15, 15-20, or greater than 20. The methylation state of sequence reads can form a methylation pattern. The DNA molecule may be one DNA molecule of a plurality of DNA molecules and method 1230 may be performed on the plurality of DNA molecules. Methylation patterns can take many forms. For example, the pattern can be N (eg, 2, 3, 4, etc.) methylated sites followed by N unmethylated sites, or vice versa. Such methylation changes may indicate junctions. The number of contiguous sites that are methylated may differ from the number of contiguous sites that are unmethylated.

At block 1234, the methylation pattern may be slid over one or more reference patterns corresponding to chimeric molecules having two portions from two parts of the reference human genome. The reference pattern can act as a filter to identify matching patterns that indicate junctions. The number of sites matching the reference pattern can be tracked to track the matching positions corresponding to the maximum number of matching sites (ie, the number whose methylation status matches the reference pattern). The two portions of the reference human genome can be discontinuous portions of the reference human genome. The two portions of the reference human genome may be separated by 1 kb, 5 kb, 10 kb, 100 kb, 1 Mb, 5 Mb, or 10 Mb or more. The two parts may be derived from two different chromosomal arms or chromosomes. One or more reference patterns may include changes between methylated and unmethylated states.

At block 1236, matching positions may be identified between the methylation pattern and the first of the one or more reference patterns. A matching position can identify a junction between two parts of the reference human genome in the sequence read. A matched position may correspond to the maximum of the overlap function between the reference pattern and the methylation pattern. A duplicate function can use multiple reference patterns. The output can be the maximum value of an aggregation function (ie each reference pattern contributes to the output value) or a single maximum value identified over the reference patterns.

At block 1238, the junction may be output as the location of the gene fusion in the chimeric molecule. The location of gene fusions can be compared to reference locations of gene fusions for various disorders or diseases, including cancer. An organism from which a biological sample is obtained can be treated for a disorder or disease.

The matching positions can be output to the alignment function. The location of gene fusions can be refined. Refining the location of the gene fusion can include aligning the first location of the sequence read to a first portion of the reference human genome. The first location can be in front of the joint. Refining the location of the gene fusion can include aligning the second location of the sequence read to a second portion of the reference human genome. The second location can be after the joint. The first portion of the reference human genome may be separated from the second portion of the human reference genome by at least 1 kb. For example, the first portion of the reference human genome and the second portion of the human reference genome are 1.0-1.5 kb, 1.5-2.0 kb, 2.0-2.5 kb, 2.5-3. They may be 0 kb, 3-5 kb, or more than 5 kb apart.

The junctions of multiple chimeric molecules can be compared to each other to confirm the location of gene fusions.

VIII. Conclusion We have developed an efficient approach to predict the level of base modifications (eg, methylation) of nucleic acids at single base resolution. This new approach implements a new scheme to simultaneously capture the polymerase dynamics surrounding the investigated base, sequence context, and strand information. Such new transformations of kinetics have allowed us to identify and model the subtle interruptions that occur in kinetic pulses. Compared to previous methods using IPD alone, the new approach present in this patent application greatly improved the resolution and accuracy of methylation analysis. This new scheme has other purposes, e.g. ), 8oxoG (7,8-dihydro-8-oxoguanine), 8oxoA (7,8-dihydro-8-oxoadenine) and other forms of base modification as well as the detection of DNA damage. In another embodiment, this new scheme (eg, kinetic transformation similar to 2D digital matrices present in this application) can be used for base modification analysis using nanopore sequencing systems.

This implementation of methylation detection can be used with nucleic acid samples from different sources, e.g., cellular nucleic acids, nucleic acids from environmental sampling (e.g., cellular contaminants), nucleic acids from pathogens (e.g., bacteria, and fungi), and cfDNA in the plasma of pregnant women. This will open up many new possibilities for genomic research and molecular diagnostics, such as noninvasive prenatal testing, cancer detection, and transplantation surveillance. In the case of cfDNA-based non-invasive prenatal diagnosis, this new invention allows diagnostic copy number abnormalities, size, mutations, fragment ends, and bases of each molecule to be detected without experimental transformations prior to PCR and sequencing. Modifications can now be used simultaneously, thus increasing sensitivity. Imbalances in methylation levels between haplotypes can be detected using the methods described herein. Such imbalances can indicate DNA molecules (eg, extracted from a disorder, such as cancer cells isolated from the blood of cancer patients) or the origin of the disorder.

IX. Example System FIG. 124 illustrates a measurement system 12400 according to one embodiment of the invention. The system shown includes a sample 12405, such as a DNA molecule in a sample holder 12410, and the sample 12405 can be brought into contact with an assay 12408 to provide a physical characteristic 12415 signal. An example of a sample holder can be a flow cell containing a tube in which assay probes and/or primers, or droplets travel (along with assay-containing droplets). A physical characteristic 12415 (eg, fluorescence intensity, voltage, or current) from the sample is detected by detector 12420 . Detector 12402 can take measurements at intervals (eg, periodic intervals) to obtain data points that make up the data signal. In one embodiment, the analog-to-digital converter converts the analog signal from the detector to digital form multiple times. Sample holder 12401 and detector 12402 can form an assay device, eg, a sequencing device that performs sequencing according to embodiments described herein. Data signal 12425 is transmitted from detector 12402 to logic system 12403 . Data signal 12425 may be stored in local memory 12435 , external memory 12404 , or storage device 12445 .

Logic system 12403 may be or include a computer system, ASIC, microprocessor, or the like. It may also include or be coupled to a display (eg, monitor, LED display, etc.) and user input devices (eg, mouse, keyboard, buttons, etc.). Logic system 12403 and other components may be part of a stand-alone or networked computer system, or directly into a device (e.g., a sequencing device) that includes detector 12402 and/or sample holder 12401. It can be attached or built-in. Logic system 12403 may also include software executing on processor 12405 . Logic system 12403 may include a computer readable medium storing instructions for controlling system 12400 to perform any of the methods described herein. For example, logic system 12403 can provide commands to systems including sample holder 12401 such that sequencing or other physical manipulations are performed. Such physical manipulations can be performed in a particular order, eg, reagents are added and removed in a particular order. Such physical manipulations can be performed by robotic systems, including, for example, robotic arms, such that they can be used to obtain samples and perform assays.

Any of the computer systems mentioned herein may utilize any suitable number of subsystems. An example of such a subsystem is shown in FIG. 125 of computer system 10 . In some embodiments, a computer system includes a single computer device, and a subsystem may be a component of the computer device. In other embodiments, the computer system may include multiple computer devices, each of which is a subsystem and includes internal components. Computer systems may include desktop and laptop computers, tablets, mobile phones, and cloud-based systems.

The subsystems shown in FIG. 125 are interconnected via system bus 75 . Additional subsystems are shown, such as a printer 74, keyboard 78, storage device(s) 79, monitor 76 (e.g., display screen such as LEDs) connected to display adapter 82, and others. Peripherals and input/output (I/O) devices that couple to the O controller 71 are known in the art, such as input/output (I/O) ports 77 (eg, USB, FireWire®). It may be connected to the computer system by any number of means, such as a wide area network such as the Internet, mouse input device, using I/O port 77 or external interface 81 (eg, Ethernet, Wi-Fi, etc.). Alternatively, the scanner may be connected to computer system 10. Interconnection via system bus 75 allows central processor 73 to communicate with each subsystem, system memory 72 or storage device(s) 79 (e.g., A system memory 72 and/or storage device(s) 79 is a computer readable storage device(s) 79 that enables execution of multiple instructions from a hard drive or a fixed disk such as an optical disk) and information exchange between subsystems. Another subsystem is a data collection device 85 such as cameras, microphones and accelerometers, and the like.Any of the data referred to herein It may be output from one component to another component and may be output to the user.

A computer system may be a plurality of identical components connected together, for example, by an external interface 81, by an internal interface, or via storage devices that may be connected or removed from one component to another. or subsystems. In some embodiments, computer systems, subsystems, or devices can communicate over a network. In such an example, one computer can be considered a client and another computer a server, and each can be part of the same computer system. Clients and servers may each include multiple systems, subsystems, or components.

Aspects of an embodiment can be implemented in computer software in the form of control logic, using hardware circuits (e.g., application specific integrated circuits or field programmable gate arrays), and/or in a modular or integrated fashion with a general purpose programmable processor. can be implemented using As used herein, a processor may include a single-core processor, a multi-core processor on the same integrated chip, or multiple processing units on a single circuit board or networked together, as well as dedicated hardware. can be done. Based on this disclosure and the teachings provided herein, one of ordinary skill in the art will be able to implement other methods and/or implementations of the embodiments of the present invention using hardware and combinations of hardware and software. or will know and understand how.

Any of the software components or functions described in this application may be written in any suitable computer language, such as Java, C, C++, C#, Objective-C, Swift, etc., or It may also be implemented as software code executed by a processing device using a scripting language such as Perl or Python using object-oriented techniques. Software code may be stored as a series of instructions or commands on a computer-readable medium for storage and/or transmission. Suitable non-transitory computer readable media include random access memory (RAM), read only memory (ROM), magnetic media (such as hard drives or floppy disks), or optical media (such as compact discs (CD) or DVDs (Digital Versatile discs), or Blu-ray discs and flash memory, etc. A computer readable medium may be any combination of such storage or transmission devices.

Such programs may also be encoded and transmitted using carrier wave signals suitable for transmission over wired, optical and/or wireless networks according to various protocols, including the Internet. Accordingly, computer readable media may be produced using data signals encoded with such programs. Computer-readable media encoded with the program code may be packaged with a compatible device or provided separately from other devices (eg, via Internet download). Any such computer-readable medium may reside on or within a single computer product (eg, a hard drive, CD, or an entire computer system), or may be on or among different computer products within a system or network. May exist inside. A computer system may include a monitor, printer, or other suitable display for providing any of the results described herein to a user.

Any of the methods described herein can be implemented in whole or in part using a computer system including one or more processors that can be configured to perform the steps. Thus, an embodiment may be directed to a computer system configured to perform the steps of any of the methods described herein, with potentially different components each step or each Implement a group of steps. Although presented as numbered steps, the steps of the methods herein can be performed at the same time or at different times or in different orders. Additionally, portions of these steps can be used in conjunction with portions of other steps from other methods. Also, all or part of a step may be optional. Additionally, any of the steps of any of the methods may be performed using a system of modules, units, circuits, or other means for performing those steps.

The specific details of particular embodiments may be combined in any suitable manner without departing from the spirit and scope of embodiments of the invention. However, other embodiments of the invention may be directed to specific embodiments relating to each individual aspect, or specific combinations of these individual aspects.

The foregoing description of exemplary embodiments of the present disclosure has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the disclosure to the precise form described, and many modifications and variations are possible in light of the above teachings.

References to "a," "an," or "the" are intended to mean "one or more," unless specifically indicated to the contrary. The use of "or" is intended to mean an "inclusive or" rather than an "exclusive or," unless specifically indicated to the contrary. Reference to a "first" component does not necessarily require that a second component be provided. Further, reference to a "first" or "second" component does not limit the referenced component to a particular location unless explicitly stated. The term "based on" is intended to mean "based at least in part on."

All patents, patent applications, publications, and specifications referred to herein are incorporated by reference in their entirety for all purposes. Nothing is admitted to be prior art.
Reference Albert, T.; J. et al. (2007) Direct selection of human genomic loci by microarray hybridization. Nat. Methods, 4, 903-905.
Beckmann et al. (2014) Detecting epigenetic motifs in low coverage and metagenomics settings. BMC Bioinformatics, 15 (Suppl 9): S16.
Beaulaurier, J.; et al. (2019) Deciphering bacterial epigenomes using modern sequencing technologies. Nature Reviews Genetics, 20:157-172.
Blow, M. J. et al. (2016) The Epigenomic Landscape of Prokaryotes. PLOS Genet. , 12, e1005854.
Breiman, L.; (2001) Random Forests. Mach. Learn. , 45, 5-32.
Chan, K. C. A. et al. (2013) Noninvasive detection of cancer-associated genome-wide hypomethylation and copy number aberrations by plasma DNA bisulfite sequencing. Proc. Natl. Acad. Sci. U.S.A. S. A. , 110, 18761-8.
Clark, T. A. et al. (2013) Enhanced 5-methylcytosine detection in single-molecule, real-time sequencing via Tet1 oxidation. BMC Biol. , 11, 4.
Clark, T. A. et al. (2012) Characterization of DNA methyltransferase specifications using single-molecule, real-time DNA sequencing. Nucleic Acids Res. , 40:e29.
Eid,J. et al. (2009) Real-Time DNA Sequencing from Single Polymerase Molecules. Science 323, 133-138.
Feinberg, A.; P. and Irizarry, R.I. A. (2010) Stochastic epigenetic variation as a driving force of development, evolutionary adaptation, and disease. Proc. Natl. Acad. Sci. , 107, 1757-1764.
Feng, Z.; et al. (2013) Detecting DNA modifications from SMRT sequencing data by modeling sequence context dependence of polymerase kinetics. PLoS Comput Biol. , 9: e1002935.
Flusberg, B.; A. et al. (2010) Direct detection of DNA methylation during single-molecule, real-time sequencing. Nat. Methods, 7, 461-465.
Frommer, M.; et al. (1992) A genomic sequencing protocol that yields a positive display of 5-methylcytosine residues in individual DNA strands. Proc. Natl. Acad. Sci. , 89, 1827-1831.
Gai, W.; et al. (2018) Liver- and colon-specific DNA methylation markers in plasma for investigation of color cancers with or without liver metastases. Clin. Chem. , 64, 1239-1249.
Gouil, Q. et al. (2019) Latest techniques to study DNA methylation. Essays Biochem. 63(6):639-648.
Grunau, C.; (2001) Bisulfite genomic sequencing: systematic investigation of critical experimental parameters. Nucleic Acids Res. , 29, 65e-65.
Herman, J.; G. et al. (1996) Methylation-specific PCR: a novel PCR assay for methylation status of CpG islands. Proc. Natl. Acad. Sci. U.S.A. S. A. , 93, 9821-9826.
Jiang, P.; et al. (2014) Methy-Pipe: An Integrated Bioinformatics Pipeline for Whole Genome Bisulfite Sequencing Data Analysis. PLoS One, 9, e100360.
LeCun, Y.; et al. (1989) Backpropagation Applied to Handwritten Zip Code Recognition. Neural Comput. , 1, 541-551.
Lee, E. -J. et al. (2011) Targeted bisulfite sequencing by solution hybrid selection and massively parallel sequencing. Nucleic Acids Res. , 39, e127-e127.
Lehmann-Werman, R.; et al. (2016) Identification of tissue-specific cell death using methylation patterns of circulating DNA. Proc. Natl. Acad. Sci. , 113, E1826-E1834.
Lister, R. et al. (2009) Human DNA methylomes at base resolution show widespread epigenomic differences. Nature, 462, 315-322.
Liu, Q. et al. (2019) Detection of DNA base modifications by deep recurrent neural network on Oxford Nanopore sequencing data. Nature Commun. , 10, 2449.
Liu, Y.; et al. (2019) Bisulfite-free direct detection of 5-methylcytosine and 5-hydroxymethylcytosine at base resolution. Nat. Biotechnol. , 37, 424-429.
Lun, F. M. F. et al. (2013) Noninvasive prenatal methylomic analysis by genomewide bisulfite sequencing of maternal plasma DNA. Clin. Chem. , 59, 1583-1594.
Nattestad, M.; et al. (2018) Complex rearrangements and oncogene amplifications revealed by long-read DNA and RNA sequencing of a breast cancer cell line. Genome Res. , 28, 1126-1135.
Ng, A. Y. (2004) Feature selection, _L1 vs. L ₂ regularization, and rotational invariance. In, Twenty-first International Conference on Machine Learning-ICML '04. ACM Press, New York, New York, USA, p. 78.
Ni, P. et al. (2019) Deep Signal: detecting DNA methylation state from Nanopore sequencing reads using deep-learning. Bioinformatics, 35, 4586-4595
Okou, D. T. et al. (2007) Microarray-based genomic selection for high-throughput resequencing. Nat. Methods, 4, 907-909.
Olova, N.; et al. (2018) Comparison of whole-genome bisulfite sequencing library preparation strategies identifying sources of biases affecting DNA methylation data. Genome Biol. , 19, 33.
Robertson, K.; D. (2005) DNA methylation and human disease. Nat. Rev. Genet. , 6, 597-610.
Smith, Z.; D. and Meissner, A.; (2013) DNA methylation: roles in mammalian development. Nat. Rev. Genet. , 14, 204-20.
Schadt, E. E. et al. (2013) Modeling kinetic rate variation in third generation DNA sequencing data to detect putative modifications to DNA bases. Genome Res. , 23(1):129-41.
Sun, K. et al. (2015) Plasma DNA tissue mapping by genome-wide methylation sequencing for noninvasive prenatal, cancer, and transplantation assessments. Proc. Natl. Acad. Sci. , 112, E5503-E5512.
Suzuki, Y.; et al. (2016) AgIn: Measuring the landscape of CpG methylation of individual repetitive elements. Bioinformatics, 32, 2911-2919.
Watson, C.E. M. et al. (2019) Cas9-based enrichment and single-molecule sequencing for precision characterization of genomic duplications. Lab. Investig, 100, 135-146.
Zhang, W.; et al. (2015) Predicting genome-wide DNA methylation using methylation marks, genomic positions, and DNA regulatory elements. Genome Biol. , 16, 14.

Claims (41)

A method for detecting nucleotide modifications in a nucleic acid molecule comprising:
(a) receiving data obtained by measuring pulses of light signals corresponding to sequenced nucleotides in a sample nucleic acid molecule, from said data the following characteristics:
identification of said nucleotide for each nucleotide;
the position of said nucleotide within said sample nucleic acid molecule;
a width of said pulse corresponding to said nucleotide, and a pulse interval representing the time between said pulse corresponding to said nucleotide and a pulse corresponding to a neighboring nucleotide;
obtaining a value for
(b) creating an input data structure, the input data structure comprising a window of said nucleotides sequenced in said sample nucleic acid molecule, wherein said input data structure comprises: , with the following characteristics:
said identification of said nucleotide;
position of said nucleotide relative to a target position within said window;
a width of said pulse corresponding to said nucleotide; and said pulse interval;
creating, including;
(c) inputting said input data structure into a model, said model:
receiving a first plurality of first data structures, wherein each first data structure of said first plurality of first data structures corresponds to a respective nucleic acid molecule of a plurality of first nucleic acid molecules; wherein each of said first nucleic acid molecules is sequenced by measuring pulses of said light signal corresponding to said nucleotides, said modifications corresponding to respective windows of nucleotides sequenced in each first having known first states of nucleotides at target positions in each window of the nucleic acid molecule of each first data structure, each first data structure containing a value for the same property as said input data structure;
storing a plurality of first training samples, each indicative of one of said first plurality of first data structures and said first state of said nucleotide at said target position; storing, including a label of 1; and
When said first plurality of first data structures is input to said model, said plurality of first training samples are used to match or not match corresponding labels of said first label. optimizing parameters of the model based on the output of the model, the output of the model specifying whether the nucleotide at the target position in the respective window has the modification; trained by, inputting; and
(d) using said model to determine whether said modification is present at a nucleotide at said target position within said window of said input data structure. the input data structure is one of a plurality of input data structures;
the sample nucleic acid molecule is one of a plurality of sample nucleic acid molecules;
the plurality of sample nucleic acid molecules are obtained from a biological sample of a subject;
each input data structure corresponding to a respective window of sequenced nucleotides in a respective sample nucleic acid molecule of the plurality of sample nucleic acid molecules;
the method comprising:
receiving the plurality of input data structures;
inputting the plurality of input data structures into the model;
2. The method of claim 1, further comprising using said model to determine whether there is a modification at a nucleotide at a target position in said respective window of each input data structure. determining that the modification is at one or more nucleotides;
3. The method of claim 2, further comprising using the presence of said modification of one or more nucleotides to assign a disorder classification. 4. The method of claim 3, wherein said disorder comprises cancer. assigning the classification of the disorder to be that the subject has the disorder ;
5. The method of claim 3 or 4, further comprising: The method of any one of claims 3-5 , wherein the number of modifications or the site of modification is used to assign the classification of the disorder. A method according to any one of claims 1 to 6, wherein said modification is methylation. The methylation is 4mC (N4-methylcytosine), 5mC (5-methylcytosine), 5hmC (5-hydroxymethylcytosine), 5fC (5-formylcytosine), 5caC (5-carboxylcytosine), 1mA (N1 -methyladenine), 3mA (N3-methyladenine), 6mA (N6-methyladenine), 7mA (N7-methyladenine), 3mC (N3-methylcytosine), 2mG (N2-methylguanine), 6mG (O6-methyl guanine), 7mG (N7-methylguanine), 3mT (N3-methylthymine), or 4mT (O4-methylthymine). 8. The method of claim 7, wherein said methylation is 5mC. 8. The method of claim 7, wherein said methylation is 6mA. wherein said modification is methylation and said method comprises:
determining the methylation status of whether the modification is present at one or more nucleotides;
using said methylation status of said one or more nucleotides to determine a clinically relevant DNA fraction, fetal methylation profile, maternal methylation profile, presence of imprinted gene regions, or tissue of origin 3. The method of claim 2, further comprising: and. 12. The method of claim 11, wherein:
The method comprises determining a tissue of origin,
A method, wherein determining the tissue of origin comprises determining whether the sample nucleic acid molecules are of fetal or maternal origin. Determining whether a sample nucleic acid molecule is of fetal or maternal origin includes:
determining the methylation level of the sample nucleic acid molecule using the methylation state of the one or more nucleotides; and
Comparing the methylation level of said sample nucleic acid molecule to a reference value
13. The method of claim 12, comprising: 14. The method of claim 13, wherein said reference value is determined from methylation levels of one or more maternal nucleic acid molecules. 14. The method of claim 13, wherein:
comparing the methylation level of the sample nucleic acid molecule to the reference value comprises determining that the methylation level of the sample nucleic acid molecule is lower than the reference value;
Determining whether the sample nucleic acid molecule is of fetal or maternal origin comprises using a comparison to determine whether the sample nucleic acid molecule is of fetal origin;
Method. 3. The method of claim 2, wherein said modification is methylation, said method further comprising:
identifying each sample nucleic acid molecule of the plurality of sample nucleic acid molecules as aligned with a region of the genome;
using said model to determine the methylation status of whether said modification is present at one or more nucleotides of each sample nucleic acid molecule of said plurality of sample nucleic acid molecules;
determining the methylation level of the region of the genome using the plurality of methylation states of the one or more nucleotides of the plurality of sample nucleic acid molecules; and
using said methylation level to determine whether a copy number abnormality is present in said region of said genome;
A method, including 17. The method of claim 16, further comprising comparing the methylation level of said region to a reference level, wherein determining whether a copy number abnormality is present in said region of the genome comprises using the comparison. described method. 18. The method of claim 17, wherein the reference level is determined using regions without copy number aberrations of the same type. 19. The method of any one of claims 16-18, wherein the region is a chromosome and the subject is a female subject pregnant with a fetus, the method further comprising:
determining that a copy number abnormality is present; and
determining that the fetus has a chromosomal aneuploidy;
A method, including 20. The method of any one of claims 2-19 , wherein each sample nucleic acid molecule of said plurality of sample nucleic acid molecules has a size greater than a cutoff size. 13. The method of any one of claims 1-12, wherein the nucleotides within the window are determined using a circular consensus sequence without aligning the sequenced nucleotides to a reference genome. 13. The method of any one of claims 1-12, wherein the nucleotides within the window are determined without using circular consensus sequences and without aligning the sequenced nucleotides to a reference genome. the plurality of sample nucleic acid molecules are aligned to a plurality of genomic regions;
for each genomic region of said plurality of genomic regions a number of sample nucleic acid molecules are aligned with said genomic region;
A method according to any one of claims 2 to 12 , wherein the number of sample nucleic acid molecules is greater than the cutoff number. The method of any one of claims 1-23 , wherein the model comprises a machine learning model, principal component analysis, convolutional neural network, or logistic regression. said window of nucleotides corresponding to said input data structure comprises nucleotides on a first strand of said sample nucleic acid molecule and nucleotides on a second strand of said sample nucleic acid molecule;
The input data structure further includes, for each nucleotide within the window, a strand characteristic value, the strand characteristic indicating that the nucleotide is present on either the first strand or the second strand. The method according to any one of claims 1 to 24 . the sample nucleic acid molecule is a circular DNA molecule;
cleaving the double-stranded DNA molecule using the Cas9 complex to form a cleaved double-stranded DNA molecule;
26. The method of claim 25 formed by ligating hairpin adapters to the ends of the cleaved double-stranded DNA molecules. A method according to any one of claims 1 to 26 , wherein each nucleotide within said window is enriched or filtered. Each nucleotide in said window is
Cleaving a double-stranded DNA molecule using a Cas9 complex to form a cleaved double-stranded DNA molecule, enriched by ligating hairpin adapters to the ends of said cleaved double-stranded DNA molecule or filtered by selecting double - stranded DNA molecules having sizes in a size range. The method of any one of claims 1-28 , wherein the optical signal is a fluorescent signal from a dye-labeled nucleotide. 30. Any one of claims 1-29 , wherein each window associated with said first plurality of first data structures comprises at least 4 contiguous nucleotides on the first strand of each first nucleic acid molecule. The method described in . 2. The method of claim 1, further comprising using the presence of methylation to detect the tissue origin of said sample nucleic acid molecules or to identify chimeric and hybrid DNA, wherein said sample nucleic acid molecules are A method obtained from said subject. At least some of the plurality of first nucleic acid molecules each have a first location corresponding to a first reference sequence and a second location corresponding to a second reference sequence different from the first reference sequence. 32. The method of any one of claims 1-31, comprising two locations. further comprising validating the model using a plurality of chimeric nucleic acid molecules, each having a first location corresponding to the first reference sequence and a second location corresponding to the second reference sequence; 33. The method of any one of claims 1-32, comprising: said first location having a first methylation pattern and said second location having a second methylation pattern. 34. The method of claim 32 or claim 33, wherein said first location is treated with a methylase. 35. The method of claim 34, wherein said second location corresponds to an unmethylated location of said second reference sequence. 34. The method of claim 32 or claim 33, wherein the first reference sequence is human and the second reference sequence is from a different animal. 37. The method of any one of claims 1-36, wherein said window comprises at least 3 nucleotides upstream of the target position within said window. The method of any one of claims 1-30, further comprising sequencing said sample nucleic acid molecules. 39. The method of claim 38, wherein sequencing sample nucleic acid molecules comprises measuring pulses of light signals corresponding to nucleotides in said sample nucleic acid molecules. A computer readable medium storing a plurality of instructions when executed by a computer system that causes the computer system to perform the method of any one of claims 1-30. at least one storage device;
a plurality of instructions stored in at least one storage device; and
At least one processor programmed with at least some of a plurality of instructions to perform the method of any one of claims 1-30
computer system, including;

Images